Gem analysis apparatus, system and method using neural network

Abstract

A method, system, and device for evaluating a gemstone using a gemstone imaging and evaluation device. The method includes capturing a plurality of training images of a plurality of gemstones using an image capture device having a plurality of different focal length settings, training a machine learning module using the plurality of training images, capturing a query image of the gemstones, analyzing the query image using the trained machine learning module, identifying a selected feature of the gemstones within the query image, and identifying the gemstones within the selected feature. And outputting a notification of the identified selected feature. The system and the gem imaging and evaluating device implement the method.

Description

Technical Field

[0001] This disclosure generally relates to the imaging and analysis of gemstones, and more specifically to an apparatus, system, and method for imaging gemstones and using a neural network to analyze gemstone images to detect gemstone features. Background Technology

[0002] Gemstone evaluation is a subjective process, easily influenced by biases arising from the visual examination of gemstones by human evaluators. Many gemological laboratories attempt to define four main characteristics of a gemstone: cut, color, clarity, and carat weight, also known as the 4Cs. Various other properties considered to constitute an ideal gemstone—such as a diamond—can include: polish, symmetry, optical properties, scintillation, fire, brilliance, faceting quality, weight, and various other optical, qualitative, and quantitative gradings. This is typically done through visual verification, but recently it has also been done with the aid of machines, expensive instruments, and predictive mathematical modeling. However, the results remain subjective, and the methods and techniques for obtaining objectively presentable grades and results have consistently challenged gemstone evaluators.

[0003] One example of why gem grading laboratories struggle to achieve objective gem assessment is clearly defining which proportions of a diamond truly constitute an ideal or excellent cut grade. Each laboratory has its own standards—different ranges of proportions, symmetry, and other measurements—to describe gems and classify them as Ideal, Excellent, Very Good, or some other category. Therefore, this is merely a subjective interpretation of the gem's beauty. Gem grading laboratories have failed to classify such gems or jewelry with fancy or non-round shapes, such as marquise or princess cuts. Fancy shapes only receive symmetry and polish as descriptive labels for their cut grade. The arrangement of the gem's main components—including the table, crown, pavilion, and culet—and the shapes, angles, positions, refractive indices, and proportions presents a significant challenge for gem grading laboratories. These laboratories attempt to address the subjectivity by requiring human testers, mathematical models, or machines to arrive at an objective gem grading solution. However, such objective results have consistently proven difficult to achieve. Another characteristic commonly used to describe diamonds is their luster. However, this characteristic is also difficult to define, subjective, and difficult to measure even with machines.

[0004] Natural diamonds and gemstones are many times more valuable than lab-grown diamonds and laboratory-made gems. Often, even highly trained professionals or gem grading labs find it nearly impossible to visually distinguish natural diamonds from lab-grown diamonds produced using processes such as chemical vapor deposition (CVD) or high-pressure high-temperature (HPHT). Advanced machinery and technology are required to differentiate between natural and lab-grown diamonds. However, as new technologies are applied to the manufacture of lab-grown diamonds and other gemstones, even advanced machinery and processes sometimes fail to differentiate between them. Gem grading labs must constantly keep up with new technological developments, which creates opportunities for fraud or simple, even obvious, errors—such as confusing or misrepresenting lab-grown gemstones with natural ones. Such fraud and errors can cause enormous financial and reputational damage. Furthermore, even visually distinguishing cubic zirconia (CZ) from diamonds is difficult for laypeople.

[0005] Using known techniques to examine whether gemstones set in jewelry or elsewhere have been fraudulently switched or replaced—whether with similar or inferior gemstones, imitations, synthetics, or whether they have been recut, broken, or damaged—is difficult. Current identification processes require expensive or invasive techniques, such as laser engraving on the girdle, or the application of microscopic lasers or the insertion of genetic or biological markers as bio-tracers into the gemstone. However, without high-tech equipment, well-trained professionals, or technical knowledge, such identification processes are difficult to easily re-examine.

[0006] A simple method is also needed to count gemstones, especially when the image is not in full focus, or when the lighting conditions are not ideal when the gemstones are set in jewelry or in bulk.

[0007] Therefore, there is a need for a more cost-effective, usable, and accurate method for gem analysis. It is precisely in response to these and other problems in the arts that this disclosure aims to provide a technical solution for gem imaging and related evaluation techniques, overcoming the inherent problems in previous gem evaluation systems and methods. Summary of the Invention

[0008] According to one aspect of this disclosure, a method for evaluating gemstones from gemstone images is provided. The method includes the step of receiving a training image set of multiple gemstones by a computing device. The computing device has a non-transitory computer-readable storage medium and a processor configured to execute software programs stored in the storage medium. The training image set is captured using an image capture device. The image capture device may include one or more macro lenses. The image capture device may include an illumination system. Each training image set includes multiple images of a corresponding gemstone from a plurality of gemstones captured at different focal length settings. The method also includes the step of training a machine learning algorithm using the processor. Specifically, the machine learning algorithm is trained to detect at least one gemstone feature from the one or more gemstone images.

[0009] The method further includes the steps of: receiving a query image of a gemstone at a processor; and analyzing the query image using a trained machine learning algorithm. Furthermore, the trained machine learning algorithm performs a step of identifying one or more of at least one gemstone feature based on the query image. Additionally, the method includes a step of outputting a notification of the identified one or more gemstone features. The method may also include a step of performing a task based on the result of identifying at least one gemstone feature. For example, the method may include subjective grading and analysis of at least one gemstone feature. The method may also include objective grading and analysis of at least one gemstone feature. The method may also include deductive grading and analysis of at least one gemstone feature. Furthermore, the method may include valuing the gemstone based on at least one gemstone feature. Moreover, the method may include other user objectives based on at least one gemstone feature. Alternatively, the method may include other predetermined objectives based on at least one gemstone feature. For example, the method may perform predetermined objectives and tasks.

[0010] According to another aspect, a system for evaluating gemstones from gemstone images is disclosed. The system includes an image capturing device having multiple different focal length configurations to capture query images of the gemstones and a gemstone evaluation device. The image capturing device can be a fixed device with known settings—e.g., a predetermined focal length setting. Alternatively, the image capturing device can be a mobile device, such as a handheld device. A user can move the mobile device to change its focus. Alternatively, a user can fix the settings of the mobile device and then move the mobile device in different directions—e.g., up and down—while maintaining the fixed focus. Using this single fixed focus, an image can be captured at the fixed focus. The gemstone evaluation device includes a processing unit containing a machine learning algorithm. Specifically, the machine learning algorithm is trained using a training image set of multiple gemstones. Each training image set includes multiple images of a corresponding gemstone from multiple gemstones captured at different focal length settings and using different imaging and lighting conditions. As a result, the machine learning algorithm is trained to detect at least one gemstone feature from one or more gemstone images.

[0011] Furthermore, the processing unit is configured to receive a query image of a gemstone from an image capture device and analyze the query image using a trained machine learning algorithm. Specifically, the trained machine learning algorithm is configured to identify one or more features of the gemstone from at least one gemstone feature based on the query image. The system also includes an output device configured to output a notification of the identified one or more gemstone features.

[0012] Any combination of the various embodiments and implementations disclosed herein may be used. These and other aspects and features will be understood from the following description of certain embodiments of the invention, as well as from the accompanying drawings and claims. Attached Figure Description

[0013] This application contains at least one color-drawn drawing. Upon request and payment of the necessary fees, the Patent Office will provide a published copy of this patent application with the color drawing.

[0014] Figure 1A This is a side view of a system including a gem imaging and evaluation device according to one embodiment.

[0015] Figure 1B yes Figure 1A A schematic diagram of the system.

[0016] Figure 2A yes Figure 1A The magnified side view of the system illustrates the focus of the gemstone.

[0017] Figure 2B It is a set of gem images captured by gem imaging and evaluation equipment at different focal points and distances.

[0018] Figure 2C It is a set of gem images captured at different focal points by gem imaging and evaluation equipment.

[0019] Figure 2D It is a set of images of natural and synthetic gemstones captured by gemstone imaging and evaluation equipment.

[0020] Figure 2E This is another set of images of natural and synthetic gemstones captured by gemstone imaging and evaluation equipment.

[0021] Figure 2F These are a set of images of gemstones placed in a box with a window, captured by gemstone imaging and evaluation equipment.

[0022] Figure 2G It is a set of images of a gemstone captured by a gemstone imaging and evaluation device, illuminated by a laser at different focal points.

[0023] Figure 3 It is used for Figure 1A A schematic diagram of the system's machine learning module, which includes a neural network.

[0024] Figure 4A According to this embodiment Figure 1A A flowchart of the overall system operation method.

[0025] Figure 4B This is a flowchart of the method for training the machine learning module.

[0026] Figure 4C It uses the trained machine learning module. Figure 1A A flowchart of the system operation method.

[0027] Figure 5A It is a set of images showing the sparkle of gemstones.

[0028] Figure 5B It is a set of images showing the color changes of gemstones.

[0029] Figures 5C-5D It is a group of images that display patterns on gemstones—including patterns inside the gemstone, from the gemstone, on the gemstone, or around the gemstone.

[0030] Please note that the accompanying drawings are illustrative and not necessarily drawn to scale. Detailed Implementation

[0031] Exemplary embodiments of this disclosure relate to a gemstone analysis system and method for imaging gemstones and analyzing the gemstone images using a neural network-based image processing algorithm. The system includes a gemstone imaging and evaluation apparatus comprising a camera and a processing computer. In some embodiments, one or more images of the gemstone are captured and analyzed so that the physical and optical properties of the gemstone can be identified, and the quality of the gemstone can be evaluated based on objective and deductive characteristics. In some cases, the system can be used for gemstones that are part of a jewelry item, such as a ring or necklace. The system can also be used to analyze one or more loose gemstones, or gemstones set in a gemstone holder.

[0032] Existing gem analysis systems and methodologies seek to identify gem features from one or more images focused on or within the gemstone. However, capturing precisely focused gem images is difficult without expensive, dedicated gem imaging systems operating in a controlled environment. Capturing precisely focused gem images is particularly challenging when using conventional camera equipment—such as digital cameras on mobile devices or smartphones. Variations in lighting conditions, human error, or hand tremors can cause blurry, out-of-focus images, hindering the extraction and processing of gem feature information using conventional image-based gem analysis techniques. Therefore, conventional image-based gem analysis techniques may produce erroneous or incorrect results.

[0033] According to a significant aspect, embodiments of the gemstone evaluation system and method disclosed herein are specifically configured to extract feature data and generate an objective evaluation of the gemstone using even blurred, out-of-focus images (e.g., as a result of over- or under-focus images). Specifically, the gemstone evaluation system and method provide a solution that captures a set of gemstone images—including focused, over-focused, and under-focused images—within a certain focal length setting range, and processes this set of images using one or more neural network-based algorithms trained to detect gemstone features that can be used to quantify various measures of gemstone quality. In some cases, images can be captured until the gemstone's halo, aura, or any other light or pattern emanating from the image is minimized or even reduced to zero. For example, images can be captured using the camera's automatic exposure (AE) or autofocus (AF) settings. At a certain moment, the camera setting may cause all halo, light, and color to disappear, producing a "black" image. When such a "black" image is obtained, the corresponding camera setting can be used as a "stop" setting. This "stop" setting can also be used as a distance measurement tool to determine the distance the camera has moved from a point to reach the "stop" setting point. This process allows us to measure the distance traveled, thus determining how far the light has traveled. Measuring this distance can also be used to assess the gemstone's brilliance and other characteristics.

[0034] Figure 1AThis is a side view of an exemplary gemstone evaluation system 100 for imaging and analyzing a gemstone 150 according to one embodiment. System 100 includes an image capture and gemstone evaluation device 112 for imaging and evaluating the gemstone 150. The image capture and gemstone evaluation device 112 includes an image capture assembly 120 having a lens 125. The lens 125 may include any known lens configuration or type of optics. The image capture and gemstone evaluation device 112 can be held on a bracket 113 extending vertically from a base 115. The bracket 113 can orient the device 112 at any angle. Furthermore, the bracket 113 can position the device 112 in any location. The device 112 may be a handheld device, and the bracket 113 can releasably hold the handheld device 112. In some embodiments, system 100 includes a gemstone holder 102 positioned on the base 115 and configured to hold the gemstone 150 during imaging. One or more light sources 160 may be configured to illuminate the gemstone 150 during imaging. For example, in one embodiment, the light source 160 may define an annular light source arranged around the lens 125.

[0035] Figure 1B This is a schematic diagram of an exemplary image capture and gem evaluation device 112 of system 100. As shown, network 114, external storage 116, and gem evaluation platform 118 can communicate with image capture and gem evaluation device 112. Image capture and gem evaluation device 112 includes image capture component 120, memory 122, processing unit 124, and input / output (I / O) device 126. Processing unit 124 can be any known type of processing device, such as a processor, microcontroller, or microprocessor.

[0036] The image capture and gem evaluation device 112 can be any known device that includes the image capture component 120 and can be configured to capture and process one or more images of gems as described herein. The image capture component 120 can be embodied in a camera. In some embodiments, the image capture and gem evaluation device 112 is a mobile phone, such as a smartphone, and the image capture component 120 includes a mobile phone camera. In one embodiment, the lens 125 of the image capture component 120 can include a lens from a mobile phone camera; however, alternatively or otherwise, it can include a macro lens, which can improve image quality. In an example configuration, the macro lens can be a wide-angle lens or a 35mm lens.

[0037] Memory 122 may be or may include program memory, read-only memory (ROM), random access memory (RAM), or a cloud-based storage environment. Memory 122 may store image data received from image capture component 120 for processing by processing unit 124. Processing unit 124 may access the data to be processed in memory. For example, for data stored in cloud-based memory 122, processing unit 124 may use cloud-based processing to process the data. Memory 122 and processing unit 124 may be configured in a mobile device. Processing unit 124 processes image data, as described herein, to evaluate gem 150 or to facilitate further processing via I / O device 126 using network 114, external storage 116, and external gem evaluation platform 118. I / O device 126 may include a transceiver, network communication interface, or any known communication device. I / O device 126 may be a transceiver or network communication device configured to transmit information corresponding to gem evaluation. Therefore, it should be understood that the gem evaluation method described as being executed by processing unit 124 can be similarly implemented, wholly or partially, by gem evaluation platform 118 or other computing devices communicatively coupled to processing unit 124. Furthermore, I / O device 126 can be a user interface and / or a display configured to output notifications or other such information corresponding to the evaluation of test gem 150 to the user. For example, the display can be a touchscreen configured to receive user input and display output notifications and information to the user. The display can also display a graphical user interface (GUI) configured to allow the user to interactively control system 100, such as manually initiating imaging and evaluation of test gem 150 and viewing displayed notifications and information. Such a GUI can allow the user to add, input, change, manipulate, or otherwise control data used by device 112.

[0038] For example, a notification output by device 112 can provide the location and type of various gemstone features on an actual query image of gemstone 150. Features identified in the notification may include, for example, inclusions, grains on the gemstone, polishing marks, scratches, internal patterns, external patterns, gemstone color, clarity, cut, symmetry, facets, edges, gemstone shape, and color variations of the test gemstone. For example, external patterns may be generated by phantoms, ghosting, or holograms. Furthermore, other features of the test gemstone identified in the notification may include optical properties, formed light patterns, created light patterns, and other known features of the gemstone. The location of features output in the notification can be represented by a bounding box displayed on the query image, data points including feature coordinates, or a representation of the image (e.g., a vector). Other representations of the image may include emojis or other symbols or characters. For example, using a bounding box, the segmentation of features can be displayed on the query image. The query image may also be colored to represent the features in the query image. Furthermore, notifications of feature locations or types can be transposed onto an image, such as a photograph or printout of the test gemstone 150, a line drawing, or any other such visual representation. In addition, the notification may include a numerical probability of the presence of a specific type of feature at the location of test gem 150.

[0039] Network 114 can be any known network, such as the Internet, a cellular network, or any other type of network, such as a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile network, or a wired or wireless network. External gemstone evaluation platform 118 can be any known platform operated or otherwise controlled by an entity, such as a gemstone dealer, a user grading gemstones, or an entity that evaluates gemstones to determine grading and pricing for use in sales—such as a laboratory. External storage 116 can be any known database or other known storage component operated or otherwise accessible by such an entity.

[0040] According to one significant aspect, the image capture and gemstone evaluation device 112 can be configured to capture multiple images of the gemstone with the objective lens 125 of the image capture assembly 120 set to multiple different focal length settings relative to the gemstone 150. More specifically, the image capture assembly 120 can be configured, under the control of the processing unit 124, to capture images of the gemstone when focused on or within a portion of the gemstone 150, or when over-focused and / or under-focused relative to the gemstone 150. For example, Figure 2AThis is a side view illustrating the position of the gem evaluation device relative to the gem 150 and the respective focus (fp1, fp2, ... fp8) of multiple captured images. In one example, fp1-fp2 are out of focus, fp3 is in-focus on the top of the gem 150, fp4-fp6 are overfocused, and fp7-fp8 are underfocused, meaning the focus is outside the gem 150. It should be understood that although fp4-fp6 are overfocused, different parts of the gem 150 can be in focus. Furthermore, changing the focal length setting of the image capture assembly 120 or changing the distance between the image capture assembly 120 and the gem 150 can cause the focal length settings f1-f8 to be in focus, overfocused, underfocused, or out of focus. Further examples, such as... Figure 2A As shown, an image can be captured while the image capture assembly 120 is held stationary above the gemstone 150 and focused on the topmost portion of the upright gemstone 150—for example, its table (e.g., focus fp3). Alternatively, in another example, an image can be captured from a side angle. In another alternative embodiment, an image can be captured using a 360-degree view around the gemstone. Another image can be captured while the image capture assembly 120 is focused on the bottommost portion of the gemstone 150—for example, its apex (e.g., focus fp6). Additional images can be captured when the image capture assembly 120 is focused at a specific depth within the gemstone 150 between its top and bottom ends (e.g., focus fp3, fp4, fp5). Furthermore, one or more images of the gemstone can be captured when the image capture assembly 120 is overfocused (e.g., focus fp3-fp6) and one or more images can be captured when the image capture assembly 120 is underfocused (e.g., focus fp7, fp8).

[0041] For example, Figure 2B This includes eight top-view images 210-1 to 210-8 of an upright gemstone 150—specifically a diamond—captured by image capture component 120 at eight different underfocus focal length settings. The diamond can be placed face up and illuminated in a darkroom. Alternatively, the diamond can be illuminated in a lit room under various lighting conditions. Halo, petals, blurred light returns, and pixels can then be extracted from the images. Furthermore, the color of the diamond can be extracted more easily under various imaging methods at various predetermined focal points. The color or multiple colors of the gemstone 150 can be identified, and the pattern and light distribution of the gemstone 150 can be analyzed for objective mapping and grading of the pattern, light distribution, etc. Figure 2C Here is another example of a gemstone—particularly a princess-cut diamond—under different focal length settings. Image 220-1 shows the diamond in a state of being almost overfocused but not fully focused, while image 220-2 shows the diamond in an overfocused state.

[0042] For discussion purposes and without limitation, the exemplary neural network-based image analysis algorithm is described as being performed on a set of images, wherein at least one image is captured by the image capture component 120 in over-focus, focused, and under-focus settings, respectively. It should be understood that more or fewer focused, under-focus, over-focus, or out-of-focus images may be captured and analyzed depending on the type of gemstone feature being evaluated. For example, the captured image may be a single image at a single focal length setting, such as an over-focus image, or it may be a set of images at a single focal length setting. It should also be understood that the training dataset may consist of a set of only one focal type or focal points, or a combination of images of one or more focal types. Such gemstone features may include the type of gemstone. Each gemstone has various features, such as diamond features different from those of a ruby. For example, multiple images may be captured at focused optical settings, or under-focus settings, or over-focus settings, or combinations thereof. Alternatively, the at least one image may be captured in at least one of the over-focus, focused, and under-focus settings.

[0043] In one embodiment, the processing unit 124 includes a machine learning module 154, which includes one or more supervised machine learning systems or one or more unsupervised machine learning systems. Machine learning modules may include, for example, Word2vec deep neural networks, convolutional architectures for fast feature embedding (CAFFE), artificial immune systems (AIS), artificial neural networks (ANN), convolutional neural networks (CNN), deep convolutional neural networks (DCNN), region-based convolutional neural networks (R-CNN), the You Only See Once (YOLO) method, Mask-R-CNN, deep convolutional encoder-decoder (DCED), recurrent neural networks (RNN), neural Turing machines (NTM), differential neural computers (DNC), support vector machines (SVM), deep learning neural networks (DLNN), Naive Bayes, decision trees, logistic model tree induction (LMT), NBTree classifiers, case-based modules, linear regression modules, Q-learning, temporal difference (TD), deep adversarial networks, fuzzy logic, K-nearest neighbor, clustering, random forests, rough sets, or any other known machine intelligence platform capable of supervised or unsupervised learning.

[0044] For example, such as Figure 3As shown, the machine learning module 154 includes a neural network 156 having multiple nodes or artificial neurons arranged in multiple layers, including an input layer, at least one hidden layer, and an output layer. The neural network 156 is configured to receive at least a portion of an image at the nodes of the input layer. Specifically, the neural network 156 is configured to receive and be trained from multiple training images depicting the gemstone, for example, images captured using the image capture component 120. Training involves receiving such training images and configuring the connections and connection weights between the nodes of each layer. For example, the training images may include a set of images, such as eight top-view images 210-1 to 210-8 of an upright gemstone 150 captured at eight different focal length settings, as shown. Figure 2B As shown. The training image set is preferably captured from multiple different gemstones. The training image set can be captured under different conditions and settings so that the neural network 156 can be trained to be robust to different devices, conditions, and settings. For example, different image sets may have different combinations of focal length settings, different camera positions relative to the gemstones, different camera devices and therefore different settings, different optics, different lighting conditions, and other possible differences. In addition, other image types can be used as training images, such as images without gemstones or without visible light return, which can be empty images used to train the neural network 156. Other image types may include images with automatic exposure (AE) lock, and images where the gemstone has traveled a distance greater than the light that can be received at the AE lock position. Alternative image types may also include spoof images. Such spoof images may be unrelated to gemstones or products or jewelry items containing such gemstones. By including such spoof images in the training image set, the machine learning module 154 can learn which objects to avoid when evaluating gemstones or products or jewelry items containing such gemstones.

[0045] Alternatively, training images can be enhanced by computationally altering existing training images. For example, Figure 2B The eight top-view images 210-1 to 210-8 shown can be flipped about an axis, or rotated by one or more angles, such as 45 degrees and 90 degrees. After the neural network 156 is trained with training images, the neural network 156 is configured to receive and process at least one query image of the test gem 150 from the image capture component 120. For example, the query image of the test gem 150 to be evaluated is received from the image capture component 120 at the input layer and processed by the node layer of the neural network 156. The output layer generates at least one signal indicating the recognition and classification of at least one feature of the test gem 150 determined from the query image. Other known training methods can be implemented to train the neural network 156. Alternatively, other known training methods can be implemented to train the machine learning module 154.

[0046] The captured query image of gemstone 150 can be analyzed by processing unit 124 through neural network 156 implementing neural network-based algorithms and other image processing and gemstone evaluation techniques to identify and analyze various physical characteristics of gemstone 150. Other image processing and gemstone evaluation techniques may include ranking or classifying the image or the output of neural network 146. Ranking or classification can be based on the corresponding grade of gemstone 150. For example, for the color of gemstone 150, diamonds can use grades from D to Z. Alternatively, grades from D to E can be used to classify the color of a diamond against another diamond of very similar quality. Grades D can range from D0 to D100, or D1 to D100, and grades E can range from E0 to E100, or E1 to E100. Such grades can be displayed or can be objectively graded. The algorithm configuration processing unit 124 for analyzing images of gemstones or jewelry items extracts one or more of various physical characteristics of the gemstone from the image, including but not limited to: sparkle, color, pattern, size, dimensions, symmetry, light return or performance, finish, cut grade, clarity, treatment, facets or carat weight of the gemstone, inclusions in the gemstone, scratches on the gemstone, dust or grains on the gemstone, table facet structure of the gemstone, girdle structure of the gemstone, girdle features of the gemstone, angles and height of the gemstone, pavilion depth and angle, crown height and angle, weight and color of the gemstone, surface coverage of the gemstone, identification marks or inscriptions on the gemstone, etc.

[0047] The algorithm configuration processing unit 124 for analyzing images of gemstones and jewelry extracts one or more of various physical features of the jewelry from the images, including any of the aforementioned gemstone features, identification marks or inscriptions on the metal parts securing the gemstone, the size, volume, angle, color, weight, metal mass and carat value of the metal parts, notches, scratches, dents, cracks on the metal parts or gemstone, the distance between the prongs, the placement of the prongs relative to each other and relative to the gemstone, the height and thickness of the prongs, the facet structure and angles of the prongs, the curvature angle of the prongs and their angles facing other prongs, the setting height of the gemstone relative to the prongs and other gemstones, and the placement position of the gemstone relative to the prongs. Gemstone surface area coverage measures whether the gemstone surface is too short (e.g., near the edge), too high (towards the table), the distance is correct, or whether the gemstone is missing.

[0048] The information mentioned above is exemplary and should not limit the scope of the invention. It should be clearly understood that processing unit 124 may be configured to extract any additional information required for analyzing gemstones or jewelry articles from actual gemstone images. It should also be understood that, additionally or alternatively, in some embodiments, multiple gemstones 150 may be analyzed simultaneously according to the techniques described herein.

[0049] Figure 4AThis is a flowchart of an exemplary method 400 for detecting gemstone features using an image captured by system 100 according to one embodiment. For example, method 400 is described as being implemented using processing unit 124. It should be understood that portions of method 400 and other methods disclosed herein can be performed using known custom or pre-programmed logic devices, circuits, or processors—such as programmable logic circuits (PLCs), computers, software, or other known circuits, such as ASICs or FPGAs, configured by code or logic to perform their assigned tasks. The device, circuit, or processor can be, for example, a dedicated or shared hardware device, such as a laptop, workstation, tablet, smartphone, part of a server, or dedicated hardware circuitry such as an ASIC or FPGA. The device, circuit, or processor can also be or may include a computer server, part of a server, or a computer system. The device, circuit, or processor may include a non-transitory computer-readable medium (CRM), such as read-only memory (ROM), a flash drive, or a disk drive, storing instructions that, when executed on one or more processors, cause portions of method 400 or other disclosed methods to be performed. It should be noted that in other embodiments, the order of operations may vary, and some operations may be omitted. The device, circuit, or processor may also include a user interface equipped with a touchscreen, such as the touchscreen of the image capture and gem evaluation device 112, or the touchscreen of a mobile phone, to allow computer interaction.

[0050] In step 405, the image capture component 120 of the gem evaluation device 112 receives or captures multiple training images of multiple gems. For example, the image capture component 120 can capture still images, videos, or a series of still images from a video. The image capture component 120 has multiple different focal length settings, different lighting settings, and different gem support settings. More specifically, in step 405, the image capture and gem evaluation device 112 can be configured to capture a set of images of the gems using different focal length settings, including at least one image captured when focused on the gem or a portion thereof, at least one image captured when underfocused, at least one image captured when out of focus, and at least one image captured when overfocused. Out-of-focus images can be images where features are not visible, such as when the image is completely black. Such focused, underfocused, out-of-focus, and over-focused images can be configured as a variable focal length image set.

[0051] In step 410, information is provided for each of the plurality of training images, which identifies at least one gem feature of the corresponding gem corresponding to each training image. This type of information may include one or more of the following characteristics: gem type, type of treatment performed on the gem, whether the gem is natural or synthetic, presence of gem support, gem scintillation, sparkle, brightness, color, pattern, size, shape, cut grade, dimensions, symmetry, light return or properties, pattern of returned light, pattern of light falling on or around the imaging surface, circular or other pattern of light created by an illumination device, pattern of light emitted from or from the gem—e.g., laser or other types of light—color of light, background color, foreground color, finish, cut grade, clarity, treatment, facets, girdle information, or carat weight of the gem, inclusions in the gem, scratches on the gem, dust or particles on the gem, other objects in the image, table facet structure of the gem, girdle structure of the gem, girdle features of the gem, angles and height of the gem, pavilion depth and angle, crown height and angle, weight and color of the gem, surface coverage of the gem, identification marks or inscriptions on the gem, and the reasons for the gem characteristics—e.g., inclusions and responses to gem characteristics—e.g., scintillation. Any other information may be included in multiple training images, such as user-defined information, predefined information, or information required to obtain a specific result.

[0052] In step 415, the neural network 156 of the machine learning module 154 is trained using multiple training images. This training may include receiving such training images at the neural network 156 and reconfiguring the connections and connection weights between nodes in each layer, such as... Figure 3 As shown.

[0053] It should be understood that a given neural network model can be trained to detect and evaluate multiple gemstone features. Similarly, a given neural network model can focus on detecting and evaluating at least one specific type of gemstone feature.

[0054] As a non-limiting practical example, an exemplary method for training a neural network model for analyzing inclusions in gemstones is further described below.

[0055] The training image set includes, for example, images of 1000 gemstones. These images may include 700 sets of images for training purposes, and 300 test images for validating the trained neural network.

[0056] In addition to one or more images of each gemstone, ground truth information about each gemstone and image is provided to the neural network 156. For example, ground truth information about a given gemstone may include a description of high-level subjective / objective characteristics, such as clarity. For instance, a clarity grade (e.g., FL (Flawless), IF (Internally Flawless), VVS1 (Very Slightly Inclusions), etc.) is provided for each gemstone. Preferably, to adequately train the neural network 156 to detect each feature of interest, the training image set is curated to have statistically significant sample size and variability. For example, images from 100 VVS1 gemstones, 100 IF gemstones, 100 VVS2 gemstones, etc., can be used to train the neural network 156 to perform inclusion detection.

[0057] Information regarding the true value of each gemstone can also include the location and classification of specific physical features of interest present in the image. For example, each gemstone image can be segmented to specify the location of each inclusion within the gemstone, and each inclusion can be further classified by type. The location of each inclusion can be specified by drawing a bounding box around the inclusion shown in the image, by occluding the edges of the inclusion (e.g., by drawing a boundary line around its outer edge), by providing the location using coordinates, etc. Furthermore, for each inclusion, its classification is provided, for example, specifying the color / type of the inclusion (e.g., black, white, etc.). The type of inclusion can also include, for example, cloud-like inclusions, feather-like inclusions, pinpoint inclusions, or any other known type of inclusion. Additionally, other features shown in the image that are related to the gemstone features in question (e.g., inclusions) can be similarly segmented and classified. For example, reflections or shadows caused by inclusions can be identified by location and classified by type.

[0058] True value information can similarly include the location and classification of other features of the gemstone, including, for example, scratches, polishing marks, internal patterns resulting from the formation of the gemstone's source rock, the gemstone's edges, and facets.

[0059] True value information can similarly include information describing other characteristics of the gemstone (e.g., type, grade / value, location), such as its color, color variations present in the gemstone, and any other type of feature of interest.

[0060] The true value information may also include the location and classification of other optical features displayed in the image, which are generated by the physical characteristics of the gemstone itself. For example, these optical features may include light reflected from the facets, edges, inclusions, and other such physical features of the gemstone. It should be understood that significant reflections or shadows may include those that appear on or within the gemstone in the image, as well as those that may appear on surfaces surrounding the gemstone in the image. More specifically, significant light reflections captured in the image may include unique shapes and patterns of light formed or reflected by light reflected, refracted, diffracted, or transmitted by the gemstone on surfaces surrounding the gemstone (e.g., base 115, gemstone support 126, mirrors in the imaging area, and other such surfaces). Such optical features presented in the image are also referred to herein as “halo,” “aura,” or “hologram.” Furthermore, optical features may be referred to as “petals,” or various unions of them, various intersections of them, or unions of them divided by intersections.

[0061] The ground truth information can also include the location and classification of other features displayed in the image that may not be derived from the physical characteristics of the gemstone itself. For example, these features may include image artifacts, light reflected from other objects—such as the gemstone setting, jewelry, or the surface of the base. In this way, the neural network 156 can be trained to ignore or even remove unwanted or unimportant image features and image artifacts caused by poor conditions, defective lenses, etc. It should be understood that any combination of image features and artifacts, whether significant or not, can be fed into the neural network as ground truth information for a given gemstone and / or gemstone image in one or more ways, including segmentation, classification, grading, and provision.

[0062] In addition to ground truth information about the features depicted in the image, ground truth information may also include the capture settings for each image. Image capture settings are a general term intended to refer to the camera settings, lighting settings, and arrangement of the camera, lighting, gemstone, gemstone holder, etc., when capturing a given image. For example, such image capture settings may include parameters such as the type of imaging component, focal length settings (e.g., focal length, focus, etc.), lens type, lens arrangement or settings, lighting configuration (e.g., light source location, type, intensity, wavelength), the position of the camera and / or light source relative to the gemstone (e.g., distance, angle, etc.), and the orientation of the gemstone.

[0063] Once the neural network 156 is trained with given ground truth information, upon receiving a query image from the gem 150, the trained neural network 156 can output the recognition of features displayed in the query image, determine whether the feature is an inclusion, and measure the probability value of the feature being an inclusion. Furthermore, the trained neural network 156 can output the location of any inclusions in the gem 150 and the type of inclusion detected.

[0064] For all other features in a query image that can be recognized by the trained neural network 156, the trained neural network 156 can provide the same information for a given feature. For a given feature, the trained neural network 156 can output the recognition of the given feature displayed in the query image, indicating the association of the feature with the gem 150, the probability of feature recognition, and the location of the feature within the gem 150. Using the aforementioned I / O device 126, such as a GUI, a user can increase or decrease the probability threshold to display only those feature recognitions whose recognition probability reaches or exceeds the selected threshold. For example, for the facets of the gem 150 in a query image that can be recognized by the trained neural network 156, the trained neural network 156 can output the recognition of the facets associated with the gem 150, the probability of facet recognition, and the location of the facets within the gem 150.

[0065] For the location of the trained neural network 156's output for a given feature, this location can be represented as the output from I / O device 126, such as a printed output or a graphic image on a display. The location in the output from I / O device 126 can be represented as a bounding box displayed on the printed or displayed image of gem 150, the bounding box surrounding the given feature at the corresponding location. Alternatively, the location can be represented by a set of data points output by I / O device 126, which includes the coordinates of all recognized features and other feature data in the query image. In another alternative embodiment, the location can be represented by multiple response vectors from the trained neural network 156, these vectors can be represented on a single resulting image from I / O device 126.

[0066] Furthermore, in another embodiment, I / O device 126 may transpose one or more different types of features onto an output photograph generated from the query image. Alternatively, different types of features may be transposed onto a line graph, such as a chart. The identified features detected in the query image by the trained neural network 156 may be further output using various known visual methods, such as a three-dimensional (3D) representation of the test gem 150, and the identified features may be superimposed on that 3D representation.

[0067] The output response of the trained neural network 156 used to detect the first feature can be displayed separately from the output response of the trained neural network 156 used to detect the second different feature. For example, the output response of the trained neural network 156 used for facet detection can be displayed separately from the output response of the trained neural network 156 used for inclusion detection.

[0068] The computations of the machine learning module 154, such as the processing of the trained neural network 156, can be performed on any known computing device or any combination of known computing devices. For example, some computations can be performed on a telephone, while others can be performed on one or more servers. Such a telephone or server can be... Figure 1B This is part of the external gem evaluation platform 118 shown. Various calculations from various computing devices—such as telephones and servers—can then be combined and output by I / O device 126. For example, computing devices can be connected to... Figure 1B As shown in network 114, computations from computing devices can be transmitted to I / O devices 126 via network 114.

[0069] Furthermore, the neural network 156 subsequently trained to detect gemstone features can be a pre-trained neural network that has been trained to detect other types of objects and features of such objects, such as detecting and recognizing entities—like dogs or cats—and features of such objects, including shape, color, size, and other physical aspects of the object. Alternatively, the neural network 156 can be configured to handle relatively large objects, or it can be configured to handle relatively small objects, such as gemstones. Any errors in detecting or classifying objects or features can be resolved by retraining the neural network 156 with the same images. Alternatively, retraining can be performed with new images. Thus, errors can be corrected, and the neural network 156 can be trained again to accurately detect or classify objects or features.

[0070] As mentioned above, neural network 156 can be a convolutional neural network (CNN). Alternatively, neural network 156 can be any type of deep learning neural network 156. Neural network 156 can also be a combination of one or more or different known neural networks 156 and machine learning modules 154, such as Word2vec deep neural network, convolutional architecture for fast feature embedding (CAFFE), artificial immune system (AIS), artificial neural network (ANN), convolutional neural network (CNN), deep convolutional neural network (DCNN), region-based convolutional neural network (R-CNN), You Only See Once (YOLO) method, Mask-R-CNN, deep convolutional encoder-decoder (DCED), recurrent neural network (RNN), neural Turing machine (NTM), differential neural computer (DNC), support vector machine (SVM), deep learning neural network (DLNN), Naive Bayes, decision tree, logistic model tree induction (LMT), NBTree classifier, case-based module, linear regression module, Q-learning, temporal difference (TD), deep adversarial network, fuzzy logic, K-nearest neighbors, clustering, random forest, rough set, or any other known machine intelligence platform capable of supervised or unsupervised learning. For example, any new and known types of neural networks that will emerge in the future can be used.

[0071] In one embodiment, the trained neural network 156 is initially trained to detect given physical features of the gemstone 150, such as inclusions, facets, scratches, etc. The neural network can also be trained to detect combinations of features. Some features are preferably combined for analysis, for example, inclusions and facets are combined because the position of an inclusion relative to a facet is related to how the inclusion is reflected, which affects the gemstone's appearance and light transmission properties. Similarly, facets and scratches can be combined because scratches may be mistaken for facets, which also affects the gemstone's appearance and light transmission properties.

[0072] The trained neural network 156 can then be repeatedly retrained to classify the test gemstone 150 based on specially selected gemstone features. For example, retraining can begin with images of round gemstones, then continue using images of square gemstones, and so on, repeatedly using images of other types of gemstones and other gemstone shapes. The system 100 can decompose such training images into multiple sub-images, partial images, cropped images, etc., and perform training by analyzing the occluded parts of the gemstones and larger portions of the gemstone images—including the entire image of the gemstone.

[0073] The neural network 156 can be trained to detect and classify the light response and color of test gemstones using training images of different gemstones with different light responses and colors. For example, the color of a gemstone can be obtained by extracting color from individual pixels, sets of pixels, or groups of pixels in multiple gemstone images. The system 100 can also generate a heatmap of the color of each gemstone from the pixels of the gemstone images. It should be understood that the gemstone images used to train the neural network 156 can be side views, top views, bottom views, front views, and rear views, as well as perspective views from various viewpoints of the image capture component 120.

[0074] In one embodiment, the neural network trained to detect and classify features including light response, color, etc., can be the same neural network used to detect gemstone features including inclusions, facets, etc. Alternatively, the neural network trained to detect such features—including light response and color—can be a different neural network model. To train neural network 156 to detect and classify the light response of test gemstone 150, multiple images of multiple different gemstones can be captured by image capture component 120 using prescribed color illumination. For example, 5500K white light can be used to illuminate different gemstones. Image capture component 120 can also capture multiple images of multiple different gemstones at different focal length settings to generate multiple focal image sets; for example, 1000 sets of images with different focal length settings can be used to train neural network 156.

[0075] As described above, in addition to providing multiple images to the neural network 156 for training, ground truth information associated with each image is also provided to the neural network 156 for training. For each gemstone and its associated image, such ground truth information may include the gemstone's table proportions, size, cut proportions, cut depth, crown height, girdle thickness, information about the gemstone's facets, various other dimensions and subjective information related to the gemstone, subjective attribute levels, image capture parameters, optical response of light features, start and end touch points of facet lines forming the pattern, etc.

[0076] Other dimensions and subjective information can be provided from external sources, such as a database in external storage 116 or a gem laboratory that is part of the gem evaluation platform 118. Subjective attribute ratings can include information about the gem's color, scintillation, light properties, luster, etc. Scintillation can be graded or measured using any known standard method and represented in an image-based measuring instrument. Image capture parameters can be data related to the focal length setting of the image capture component 120, the type of the image capture component 120—e.g., the type of camera used, etc.

[0077] The optical response of light features can be labeled to indicate desired properties of a gemstone, such as the light pattern emanating from the gemstone, the color of each pixel in a gemstone image, and the identification of the gemstone's shape or orientation. The color of each pixel can be determined using known computer vision techniques and can be used to create a heatmap of the gemstone's color. Computer vision techniques can also measure the cleanliness of each pixel or group of pixels. The identification of the gemstone's shape or orientation can include, for example, identifying information about each lobe of light emanating from the gemstone, where the facets of the gemstone reflect, diffract, refract, or transmit light, providing the appearance of petals in certain images. The identification of the gemstone's shape can also include information indicating differences between different petals.

[0078] The recognition information provided to the neural network 156 can also associate the recognized pattern or image features with features corresponding to the gemstone. In addition to the various features of the gemstone mentioned above, other features may include the gemstone's fluorescence, phosphorescence, milky white color, turbidity, and dullness; whether the gemstone is natural or formed through CVD or other manufacturing processes; cut type; grade; various light formations; and other information about the gemstone's characteristics. For example, system 100 can use defined rules that associate light patterns with gemstone features. The concentration of light patterns can indicate, for example, features of the gemstone, such as the gemstone's cut or proportions.

[0079] Furthermore, the neural network 156 can be trained as a function of illumination and variable lighting conditions. For example, the illumination of the gemstone can come from an LED. Alternatively, the illumination can come from a laser. In another alternative embodiment, the illumination can come from ultraviolet (UV) light. In yet another alternative embodiment, the illumination can come from infrared (IR) light. Additionally, the illumination can come from a fluorescent tube. Alternatively, the illumination can be a natural light source, such as sunlight. The frequency and intensity of the LED light or laser can be varied, indicating directionality and the type of light, such as whether the light is diffuse or focused, and whether the light has a single color or multiple colors.

[0080] Furthermore, the neural network 156 can be trained using image sets and related information to remove unwanted features from images produced by insufficient or unsatisfactory imaging materials—such as defective lenses—which may cause halo effects, dark spots or shadows, or recurring halo displays. Using the trained neural network 156, images can be modified to remove such halo effects, as well as rainbow or spectral effects, image distortion, etc.

[0081] Now back Figure 4AAs described above, steps 405-415 of method 400 are for the training phase, where training images and ground truth information are provided to machine learning module 154 to train one or more neural networks 156. Steps 420-445 of method 400, further described herein, are for the testing or application phase, where a system 100 with trained neural networks 156 is used to capture and analyze images of a “test” gem 150 to detect features of interest in the test gem. Such steps 405-415 can be implemented as follows: Figure 4B The method 450 shown is performed, in which one or more neural networks 156 are trained. The resulting trained neural network 156 can be used with any known device having a processor configured to receive and evaluate one or more images of the gem 150, and optionally, to capture the images using an image capture component 120. Therefore, the trained neural network 156 can be stored and used in a smartphone or any other known computing device—such as… Figure 1B The image capture and gem evaluation device 112 shown operates within it. Alternatively, the trained neural network 156 can be stored and used on an external gem evaluation platform 118 (such as...). Figure 1B It runs in (as shown).

[0082] Furthermore, such steps 420-445 can be performed as follows: Figure 4C The method 460 shown is performed where any known device having an image capture component 120 and configured to image and evaluate the gem 150 can utilize an externally provided trained neural network 156. Therefore, through... Figure 4B Method 450 in the text describes how training neural networks can be done using... Figure 4C The imaging and evaluation of gemstone 150 by method 460 are performed separately and independently.

[0083] like Figure 1A As shown, in step 420, the gemstone 150 (i.e., the test gemstone) is positioned. For example, in step 420, the gemstone 150 is placed within the gemstone holder 102. Alternatively, the gemstone 150 may be positioned on or inside the gemstone holder 102, on a work surface, or on any surface, such as a person's hand. Then, in step 425, one or more light sources 160—such as light-emitting diodes (LEDs), incandescent lamps, lasers, UV lamps, IR lamps, tube lights, or the sun—are used to illuminate the test gemstone 150. Such illumination may be directed from above, below, or to the side of the test gemstone 150, depending on the specific analytical method being performed. The illumination may come from 360 degrees around the gemstone. Alternatively, the illumination may come from multiple locations. Furthermore, the illumination may come from various types of light.

[0084] In one embodiment, it is preferred that the primary illumination source is directed onto the test gemstone 150 from above, for example, using known halo illumination techniques. It is also preferred to illuminate the test gemstone 150 with white light. However, certain illumination conditions, such as natural sunlight, can enhance the presentation of certain features in an image or obscure them. Therefore, in some embodiments, a variable focal length image set can be captured for each of several different illumination conditions, resulting in an image set that can be used to better identify the gemstone and its features. In this configuration, system 100 may include various types of light emitters, such as LEDs, configured to emit light with different properties, such that one or more types of light with different wavelengths, colors, spectra, and intensities can be directed onto the gemstone. Furthermore, the intensity, direction, or multiple directions of the light illuminating the gemstone can be varied to create different illumination conditions.

[0085] In step 430, one or more images of the test gemstone 150 are captured by the image capture and gemstone evaluation device 112. In the exemplary embodiment described herein, a set of variable focal length images of the test gemstone is captured for further processing. However, it should be understood that, as a result of the training phase, the trained neural network algorithm is capable of detecting significant gemstone features from at least one image. Similarly, in one embodiment, multiple sets of variable focal length images may be captured in step 430, for example, under different lighting conditions, for analysis.

[0086] In step 435, processing unit 124 processes the variable focal length image set to identify one or more physical and optical properties of the test gemstone 150. More specifically, processing unit 124, particularly including a machine learning module 154 comprising one or more neural network-based models 156, such as... Figure 3 As shown, the model is trained to identify one or more gemstone features of interest from image data contained in a variable focal length image set. The neural network-based model can generate vector embeddings representing the features of the test gemstone 150 detected within its programming and training capabilities. Therefore, the processing unit 124 can be configured to analyze images captured under different lighting conditions and different focal length settings using one or more neural network-based models to identify gemstone features as a function of light or focal length setting. In step 440, the processing unit 124 identifies selected features of the test gemstone 150 in the query image. Then, the method performs a task in step 445 using the identified selected features. For example, Figure 1BThe input / output device 126 shown outputs a notification of the selected features identified by the test gem 150 in step 445. Alternatively, this task can perform use cases as described below, such as matching the color of the test gem with the color of another gem—e.g., a color entered by the user. The color of the other gem can be received, for example, via an application or website, or from any other known type of method or platform. Another task could be to use the output to verify the authenticity of the gem, for example, checking whether the same gem is identical to a previously captured or stored gem image in a database. The input / output device 126 can output, for example, a TRUE or FALSE response for verification. In another example, the input / output device 126 can output a probability, such as a similarity percentage.

[0087] refer to Figure 3 Neural networks 156 can implement deep neural network models and can include, for example, tens of millions of artificial neurons. The content that a given neural network model can process and therefore output, such as vector embeddings, is generated as a function of its programming and training, and is not merely a mechanical description of precise pixels in an image. Instead, the output of a neural network model is a high-level description of image features at various levels of abstraction. The ability of neural network models to generate high-level abstract representations of images makes their output representations extremely powerful in characterizing image features and therefore the features of objects depicted within the image.

[0088] In one embodiment, the neural network is trained from training images captured at multiple focal points—preferably at the same location—to learn how each gem feature of interest (e.g., a portion of the gem, the gem's structure, inclusions or other such features on or inside the gem, or light scattering) appears under various possible focal lengths and lighting conditions. Such image processing by processing unit 124 can be performed by refocusing the image capture and gem evaluation device 112, by recreating an image of the original object by running an algorithm, or by simply determining the patterns or color patterns that constitute any inclusion, internal or external feature, gem structure, or non-gem features—such as dust, metal, plastic, etc. When inclusions or features are not visible or identifiable, for example, if there are reflections or shadows caused by reflection, processing unit 124 can infer where the original inclusion might be located based on the type of gem and its structure and lighting conditions. Out-of-focus inclusions, especially those through which light passes or around dark inclusions, produce halos, glows, light patterns, and pixel patterns that the neural network of machine learning module 154 is trained to recognize. The neural network can also be trained to recognize any external image of various objects reflected onto the gemstone, the gemstone's structure, any inclusions, and features such as the appearance of the gemstone 150 if it is internally reflected in the image or reflected on the gemstone, or if it blocks light from entering the gemstone 150.

[0089] In some embodiments, a gemstone holder 102 is provided for accommodating or supporting one or more test gemstones 150 during inspection, storage, or other processes. Examples of systems and methods for securing, imaging, and analyzing one or more gemstones or jewelry articles using machine vision and other techniques are shown and described in co-pending and co-assigned U.S. Nonprovisional Patent Application Serial No. 17 / 207,418, entitled “GEMSTONE CONTAINER, LIGHTING DEVICE AND IMAGING SYSTEM AND METHOD”—filed March 19, 2021, the entire contents of which are incorporated herein by reference.

[0090] As described above, the lens 125 of the image capture assembly 120 may include a macro lens. The term macro lens is used herein as a general term. Other devices or equipment, such as more complex devices, may be used. For example, microscopes or other devices, such as electron microscopes, holographic lenses, wide-angle lenses, bipolar lenses, multipolar lenses, polarizers, filters, various types of coatings, optical coatings, multi-lenses, additional versions, or any known type of optics, optical devices, or equipment, may be used.

[0091] The image capture component 120 can automatically or via a software-implemented application perform zooming, multi-focus locking, optical changes, and optical tracking. The image capture component may include a laser, lidar, or other distance measurement system. The image capture component can be configured to allow external devices or smartphones with the instruments and lenses described herein to take distance readings and input these readings for integration into the gem evaluation system 100 and its training and testing methods.

[0092] System 100 may include a gemstone holder 102 or base 115, which includes a movable or foldable reflective surface that can be angled to highlight specific portions of the gemstone 150 in order to capture an image of the gemstone at certain angles. Light source 160 may include fixed or movable lamps that can be configured to focus emitted light to a single point or diffuse emitted light as needed to cover the gemstone 150. The gemstone holder 102 or base 115 may also be configured to reduce external, ambient, or surrounding light that may interfere with readings. The gemstone holder may have one or more shapes and profiles, and a flat, plate-like surface configured to allow the imaging device to contact or nearly contact the surface of the gemstone 150 or gemstone holder 102.

[0093] System 100 can be configured to allow light to enter and receive returning light to send an image of gemstone 150 to the aforementioned imaging device. Processing unit 154 can segment the image, capture the segmented images using an image capture component, or merge the segmented images. The resulting processed image can be sent to other devices or processors configured for further image processing. Light source 160 can include light of different types, properties, shapes, colors, sizes, speeds, wavelengths, intensities, directions, and illumination ranges. Lens 125 can be a simple lens, with light source 160 and imaging component 120 separate from the lens. Lens 125 can be a single lens or lens system with a single or multiple focal points, configured to focus on one or more portions of gemstone 150. The lens can have different positions and orientations configured to image gemstone 150. The lens, camera, and other components of system 100 can be manually, semi-automatically, or fully automatically controlled and adjusted by a processor.

[0094] The illumination of the gemstone can be enhanced by using reflectors or mirrors inside or around the lens system. Alternatively, reflectors or mirrors can be placed in one or more areas, at a distance, or in a direction to guide light to a given location, which can be a single point or multiple points. The light can be diffused, clustered, or patterned, propagating light at various angles, colors, or patterns. The light can be controlled to flash in a patterned or random manner and can also be adjusted to a shutter or image-capturing actuator, such as a button pressed by the user. Lens 125 can be a clip-on lens, an add-on lens, a built-in lens, or other known types of lenses. More than one lens of the same or different types can be used and can be placed around the gemstone 150 in more than one location.

[0095] In one embodiment, a bracket can be used as a gemstone holder 102. This bracket can be automated, semi-automated, semi-automatic, or manually controlled, and can also be used as a bracket for imaging and lighting equipment, as well as a distance measuring instrument or measuring device if desired. One or more known distance measuring devices or mechanisms can be implemented to measure or determine the travel height or distance starting from, for example, the top of the gemstone 150. Such distance measuring devices or mechanisms can allow for maintaining the position of one or more distance measuring devices or mechanisms or multi-directional movement. The starting point of the measurement can be considered in the case where the gemstone holder 102 is a container or box holding the gemstone, such as the height and distance from the top surface of the container to the actual surface of the gemstone inside the container, including the thickness of the lid or viewing window. A gemstone holder can also be considered when measuring or determining one or more distances, heights, and widths of the gemstone 150. The user can input a measurement of the size of the gemstone 150, the container holding the gemstone 150, or any known object. The measurement can also be the distance or distribution of the container or other object relative to the size of the gemstone 150.

[0096] The properties of the container or box can also be used to train a neural network to identify and classify gems in such boxes. For example... Figure 2F As shown, images 250-1 and 250-2 illustrate a group of small diamonds in a black box with an open glass window. Image 250-1 is in focus, while image 250-2 is out of focus. The reaction of the glass window surface differs from that of a plastic window or an exposed gemstone, due to the refractive index of the window and its cleanliness or oiliness. By using different box and window materials, as well as different focal length settings, a neural network 156 can be trained using a training image set with different image characteristics.

[0097] System 100 also includes a light source 160. The light source 160 may include a light emitter, such as an LED or any other known type of light-emitting device for illuminating the top or sides of the gemstone 150. Alternatively, the light source may be exposed on the top surface of the base 115 or otherwise configured to illuminate the gemstone 150 from below through the top surface of the base 115. For example, in one embodiment, the LED is exposed through an opening in the base 115. By further example, the LED may be embedded below the top surface of the base 115 and may emit light through the top surface of the base 115.

[0098] In some embodiments, the light source 160 can be configured to emit light having one or more of a variety of possible colors and wavelengths. In some embodiments, the light source 160 has an adjustable intensity. Various characteristics of the light emitted by the light emitters (e.g., LEDs and lasers) constituting the light source 160, such as color, intensity, and other characteristics, individually or collectively, as well as the focal length setting of the light emitters, can be controlled. Such characteristics of the emitted light can be selectively controlled by a controller via embedded electronic circuitry that powers and connects the controller to the emitter. For example, as Figure 2G As shown, images 260-1 to 260-3 can be obtained from a diamond placed in an opening of a cylinder. Alternatively, the diamond can be placed in any known object, such as an opening in a gemstone holder. Optionally, prongs can hold the gemstone in place. When light moves back and forth from above, below, or to the side of the gemstone, it produces an effect similar to... Figure 2G The light pattern shown is an example of a pattern that can be imaged, saved, and verified later. For instance, the pattern can be stored in a database. This database can... Figure 1B The image is stored in memory 122 or external memory 116. A focused diamond image 260-1 is obtained using a white LED ring light. Alternatively, a laser, such as a red laser or any other known color laser, can be used, emitted from below, above, or to the side of the diamond, to obtain an image 260-2 of the diamond.

[0099] Light can be emitted by two or more light sources, which can be the same or different. For example, the light sources can have varying properties. The light sources can also have varying intensities, varying wavelengths, or varying frequencies. Such light sources can produce specific patterns or features from a gemstone, highlighting these specific patterns. Alternatively, some illumination may be canceled or enhanced due to cross-illumination, as in three-dimensional (3D) images, providing a viewable or extractable type of illumination output. For example, the illumination output of a gemstone can be treated like a fingerprint of the gemstone. Alternatively, the illumination can be a readable output that is easily examined by humans. Furthermore, the illumination can be readable by neural networks or any other known artificial intelligence (AI) or machine learning methods. In one embodiment, laser light emitted from or passing through a gemstone can be read on any surface, such as an intermediate surface. The intermediate surface can consist of paper, cloth, or any material or object that allows light to be visible on either side of the intermediate surface.

[0100] The light illuminating the gemstone can be laser light, LED light, or halo light. Any known imaging method can be used to remove unwanted light. Unwanted light can be additional or diffused light. Therefore, the remaining light can highlight only the gemstone's most concentrated or intense light response, allowing for a clearer observation of the light spot pattern. Furthermore, the distance, shape, or design of the light response can be imaged more clearly, making it easier for system 100 to verify the gemstone for security purposes, as described herein. By changing the light properties and the focal length setting of the light emitter as described above, images of the gemstone can be obtained and then used as a training set to train neural network 156 to identify and classify gemstones. In particular, the different light properties of images 260-1 to 260-3, such as the bright, prominent parts of the gemstone, stand out and can be detected by neural network 156 and used as the gemstone's optical fingerprint.

[0101] One or more light sources 160 may have meters or other recording devices configured to inspect or serve as fail-safe devices for controlling and managing the one or more light sources 160. The one or more light sources 160 may emit cool white or warm light of known color and wavelength. The color and wavelength of such emitted light may be specified based on the type of gemstone. The light source 160 may use a known type of lens mounted on a smartphone with an illumination system to adjust the illumination of the gemstone and its surroundings.

[0102] Alternatively or additionally, in addition to adjusting the focal length setting of the optics of the image capture assembly 120, images at different focal points can be effectively achieved by changing the distance of the image capture assembly 120 relative to the jewel 150 while maintaining the image capture assembly 120 at a predetermined focal length setting. This distance change can be achieved by altering the height of the adjustable support 113, such as... Figure 1AAs shown. Therefore, in one embodiment, the processing unit 124 of the image capture and gem evaluation device 112 can cause the image capture assembly 120 to automatically adjust its optical focus. In manual setup, the user can be guided to move the image capture assembly 120 toward and away from the gem 150. Furthermore, in some embodiments, the system 100 may include an automated support as a bracket 113 on the base 115, configured to hold the image capture assembly 120. The system 100 may also be configured to automatically move the image capture assembly 120 relative to the gem 150 using, for example, a linear actuator controlled by the processing unit 124. In another embodiment, the image capture assembly 120 may be combined with laser scanning technology or other known systems for imaging gems.

[0103] Various applications of the system 100, utilizing the machine learning module 154, can perform objective evaluations of gemstones 150 or other items such as jewelry. For example, the system 100 can distinguish between natural diamonds and lab-grown diamonds produced through chemical vapor deposition (CVD) or high-pressure high-temperature (HPHT) processes. Therefore, the system 100 can provide security in determining whether a gemstone 150 is genuine and natural or man-made. Furthermore, the luster of the gemstone 150 can be objectively determined by using the system 100 to find pixel distribution patterns, pixel distribution per square unit of measured length in the image, pixel color, pixel brightness, and various other features of pixels in the query image. Through such calculations, the luster of a gemstone can be objectively determined, rather than relying on the subjective judgment of a gem dealer.

[0104] In additional applications, system 100 can objectively determine the scintillation or sparkle, color, pattern, size, dimensions, symmetry, light return or performance, finish, cut grade, clarity, treatment, facets, or carat weight of gemstone 150. System 100 can also distinguish or differentiate gemstone 150 from gemstone setting 150, the setting of gemstone 150, and other gemstones.

[0105] Evaluation of flicker

[0106] The machine learning module 154 can also be trained to objectively measure the shimmer of the test gemstone 150. The shimmer of a gemstone is the flash or glint of monochromatic or polychromatic—e.g., red, green, blue, or other colors—light appearing at various locations on the gemstone, and these flashes of light have various shapes and sizes. Once trained, the machine learning module 154 can objectively measure the color, position, shape, and size of these flashes of light. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different shimmers. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images with different shimmers. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0107] Each training image is associated with recognition information that represents the shimmer of each different gemstone at each different angle. For example, Figure 5A Images 501-1 to 501-4 show the shimmer of different gemstones at various angles, with annotations identifying salient features in the images corresponding to the shimmer. Image 501-1 shows an out-of-focus image of a gemstone, displaying the aperture or halo, and the flash or shimmer. Image 501-2 shows an over-focus image of a gemstone, showing the internal colors at the locations of white and colored flashes. Image 501-3 shows an image of a focused gemstone, with the flash, the color distribution of the flash, and the thickness, size, and travel distance of the flash. Image 501-4 shows an approximate field or circle depicting a white halo around the gemstone, with arrows indicating measurable distances from the edge of the gemstone. Figure 5A In the example shown, in image 501-4, the arrows indicate the measurable distance from the edge of the gemstone. The identification information can be a numerical value measuring the shimmer of each different gemstone at each different angle. This value can be a single value representing the color of the shimmer, such as "0" for black, "1" for red, "2" for orange, etc., according to a predetermined color coding model. Alternatively, the value can be an N-tuple of multiple numbers representing the colors of the shimmer. For example, a three-tuple or triplet of numbers, such as (0, 100, 255), can represent a color according to a red-green-blue (RGB) color model. In another embodiment, the identification information can be a letter grade, such as "A", "B", etc., corresponding to the degree of shimmer.

[0108] Once the machine learning module 154 is trained to evaluate the shimmer of different gems at different angles, the system 100 can be activated to illuminate the test gem 150, such as Figure 1A As shown, the image capture component 120 captures at least one query image of the test gem 150. The trained machine learning module 154 then processes the at least one query image to obtain and output a value that objectively measures the shimmer of the test gem 150. For example, the output value could be a single digit, a tuple of digits, or an alphabetical order as described above. The output value could be the distance traveled from a given point on the gem 150—e.g., the center, geometric center, mathematical center, or edge of the gem 150—to the tip or end of the shimmer. Alternatively, the output value could also be the thickness of the shimmer or a measured total area. Such an output value could be a relative measurement of the area of ​​the gem 150—e.g., the total area of ​​the gem 150 measured in pixels per square inch.

[0109] Furthermore, the halo, aura, or other features of the gemstone 150 can be measured, and the color of such halos, auras, or other features can be extracted. The union, intersection, shape, and size of the halos, auras, or other features can all be quantified, extracted, and calculated. Additionally, models can be obtained to derive the objective values, intensity, and other measurements of the gemstone. These models can be mathematical or statistical. The model can be based on the shape of the gemstone 150. For example, the model can be based on the round or marquise shape of the gemstone 150. Alternatively, the model can be based on surface area, the face-up surface area, or other combinations of the gemstone 150 feature categories.

[0110] The trained machine learning module 154 can also assist in the cutting process of rough gemstones. By measuring the scintillation of rough gemstones, system 100 allows gem artisans to determine how and where to place facets and at what angles to improve the scintillation during the cutting process. For example, by measuring the scintillation of rough gemstones, system 100 can determine the location and manner of the junctions of the internal and atomic structures of the rough gemstone. These internal structures and junctions reflect, refract, reflect, and dissipate incident light, causing the gemstone's scintillation. Therefore, as the gemstone is cut, the objective measurement of its scintillation by system 100 can improve the scintillation to achieve the best or optimal light return and scintillation of the final cut gemstone.

[0111] Evaluate color

[0112] Machine learning module 154 can also be trained to objectively measure the color of test gemstone 150. Machine learning module 154 can also be trained to find inclusions in test gemstone 150. Furthermore, machine learning module 154 can be trained to objectively measure the milky white color of test gemstone 150. Such milky white color affects the color or clarity of test gemstone 150. Moreover, such milky white color may be caused by inclusions found in test gemstone 150. Milky white color in gemstones (such as diamonds) can be assessed by analyzing any differences in the foreground or background of the gemstone. Milky white color can also be analyzed by comparing the color, hue, luster, paleness, etc., of the gemstone as a whole or each facet of the gemstone. Furthermore, milky white color can be analyzed by the blackness, whiteness, pixelation, or color amount associated with a given facet. Alternatively, milky white color can be analyzed for the intersection of facets or the union of facets with patterns from the foreground, background, or other facets.

[0113] In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different colors. Alternatively, the image capture component 120 can repeatedly control the operation of the lens 125 or the light source 160 to capture multiple training images from multiple gemstones. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images of gemstones with different colors. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0114] To obtain such training images, the gemstone can be placed on its side. The gemstone can also be placed on or inside a color chart. Alternatively, the gemstone can be placed in a gemstone box. Furthermore, the gemstone can be placed on a gemstone holder 102 or on a jewelry item. Moreover, the gemstone can be manually secured with tweezers or in any other known manner. Additionally, the gemstone can be placed face up or face down. Additionally, the gemstone can be placed facing a host stone or sample gemstone for reference. The lighting conditions for each gemstone can also be varied. Furthermore, multiple images of the gemstone can be captured under variable focal length and lighting conditions. In this way, the image capture component 120 can capture multiple gemstone images as training images for gemstones with different colors.

[0115] Each training image is associated with recognition information that represents the color of each different gemstone at each different angle, at each different focal point, and under different lighting conditions. Figure 5BImages 502-1 to 502-4 show the different colors of various gemstones at different angles. Image 502-1 shows a front-facing, top-view view of a pair of gemstones. Image 502-2 shows a front-facing, top-view view of another pair of gemstones. Image 502-3 shows a front-facing, top-view view of a third pair of gemstones. Image 502-4 shows a side view of a fourth pair of gemstones. For example, the color of a gemstone may change under fluorescent light compared to when it is illuminated in sunlight. Furthermore, hue, value, brightness, and other color-representative characteristics can have associated identification information. Additionally, the identification information can specify various colors within the gemstone, such as whether the gemstone has more than one color. The identification information can be a numerical value that measures the color of each different gemstone at each different angle. This numerical value can be a single value representing the color of the gemstone, such as "0" representing black, "1" representing red, "2" representing orange, etc., according to a predetermined color coding model. Alternatively, the numerical value can be an N-tuple of multiple numbers representing the colors of the gemstone. For example, a three-tuple or triplet of numbers, such as (0, 100, 255), can represent a color according to the Red-Green-Blue (RGB) color model. Alternatively, hexadecimal codes or any other known model can be used to specify numbers that define gemstone colors within an objective and narrow range. Such values ​​can be used to find or match a given gemstone with... Other gemstones with similar properties or colors .

[0116] Once the machine learning module 154 is trained to evaluate the color of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as... Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the color of the test gem 150. For example, the output value could be a single digit, a tuple of digits, or a letter grade as described above.

[0117] Figure 2B The series of images shown depicts the same gemstone placed on a sponge-containing gem holder inside a box, captured from a "top view" in a variable focal length setting. It is understandable that extracting color from a single image (such as 210-1) can be difficult due to the additional color distribution produced by the fusion of black / white facets. Furthermore, any reflection or refraction of external light would create further obfuscation in such a single focused image. However, as can be seen from the series of variable focal length images captured using exemplary lighting and imaging methods, it helps to eliminate the obfuscation present in a single image and provides an effective color extraction method.

[0118] Furthermore, imaging and neural network-based analysis techniques enable the system to extract even more feature data, providing more detail about the gemstone. For example, from a set of images with variable focus, processing unit 124 can be configured to determine the "depth" of the gemstone (i.e., the distance from the culet to the flat top) or the distances between its various parts (e.g., the crown, girdle, girdle thickness, culet, and the area therein), as light highlights each part of the gemstone in a planar or other horizontal manner, even down to the pixel level, creating new patterns that only a trained neural network can detect. Figure 2B As can be seen in the image series, the black appearance of the gemstone's central table changes from relatively large in image 210-1 to smaller in image 210-4, until it is no longer visible in image 210-6. This is caused by the reflection of light within the gemstone as it strikes these different layers. Therefore, by using physical / optical / visual or other measuring devices, the processing unit can be configured to determine the distance traveled while the image is being captured, thereby determining the gemstone's depth measurement and overall structure. Alternatively, if desired, combined or stitched images can be created based on the best features collected by system 100 to produce a single image of one or more gemstones.

[0119] Evaluation pattern

[0120] The machine learning module 154 can also be trained to objectively measure the pattern of the test gemstone 150. The pattern of the gemstone can be a faceted pattern, such as a heart shape, arrowhead shape, petal shape, or other known faceted patterns of the gemstone. Faceted patterns, along with polishing, symmetry, rough stone formation, and other factors, can produce a variety of other optical patterns under various focal length settings and lighting conditions. Such other optical patterns manifest as light that is partially or completely reflected, refracted, or reflected under these conditions, such as halos, petals, and colors. Therefore, a neural network can be trained to measure the pattern of the test gemstone based on the light patterns displayed inside the gemstone in images captured under different imaging and lighting conditions. Alternatively, the measured pattern can be seen around the gemstone. Furthermore, the measured pattern can be generated by the test gemstone 150 and emanate from within the gemstone. Additionally, the measured pattern can emanate from the outside of the test gemstone 150. In one embodiment, the image capture assembly 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different patterns. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gem holder 102 can be configured to move different gems by translation or rotation, allowing the image capture assembly 120 to capture multiple training images with different patterns. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124. Alternatively, the movement of the test gem 150 relative to the image capture assembly 120 can be performed manually.

[0121] Each training image can be associated with recognition information that represents the pattern of each different gemstone at each different angle. Figures 5C-5DImages 503-1 to 503-7 are a set showing patterns on different gemstones from various angles. Image 503-1 shows a top view of a gemstone with an arrow pattern. Image 503-2 shows a top view of a gemstone with a heart-shaped pattern. Image 503-3 shows a bottom view of a gemstone with both heart and arrow patterns. Image 503-4 shows a top view of a gemstone with an internal facet pattern. Images 503-1 and 503-4 can be viewed as phantoms; the actual gemstone, such as a diamond, is two bright spots in the image. The pattern or illusion forms in mid-air, like a hologram or 3D image. Such patterns in the images can be processed to extract information about the gemstone in the image. Image 503-5 shows a top view of another gemstone with an internal facet pattern. Image 503-6 shows a bottom view of the pavilion of a gemstone with a facet pattern. Image 503-7 shows a bottom view of the pavilion of another gemstone with a facet pattern. Images 503-5, 503-6, and 503-7 can also be viewed as phantoms, with the actual gemstones, such as diamonds, being highlights in the images. The identification information can be a numerical value representing the pattern of each different gemstone at each different angle. For example, a heart shape can be represented by "1", an arrow by "2", a petal by "3", and so on. Alternatively, the identification information can be a letter value representing the pattern of each different gemstone at each different angle. For example, a heart shape can be represented by "A", an arrow by "B", a petal by "C", and so on. In another alternative embodiment, the letter value can be an alphanumeric string, such as "CM1" representing the first crown and main pattern, while "CS1" represents the first crown and secondary pattern. In yet another alternative embodiment, the alphanumeric string can specify (area + name + (number or name)), which equals the number of facets. Alternatively, the alphanumeric string can be (facet A#1 + facet B#1) representing the pattern, while (facet A#1 + facet B#1) represents the pattern. N represents a larger pattern.

[0122] These designations can also indicate the intersection or union of patterns. In some cases, the designation can indicate a negative result. For example, the designation can indicate an undesirable pattern in a gemstone, such as one resulting from a poor cut. In another example, the designation can indicate a lack of pattern or a lack of continuous pattern formation.

[0123] Identification information can be based on cut type, such as round or emerald cut, and on the number of facets and possible patterns. Furthermore, information about the facets can include their type, shape, and the total number of facets. Identification information can also be based on the imaging angle and the part of the gemstone being imaged, such as the crown, girdle, or pavilion area. If additional facets are identified during imaging, the gemstone identification information includes a notification or label indicating the additional facets. Additionally, the user can be notified of such additional facets in the output message or alert.

[0124] The numerical values ​​used to identify and represent patterns may be due to a feature of the gemstone image. For example, the image may be over-focused, in focus, under-focused, or out of focus. These values ​​may also be due to any halos, rings, petals, facets, and their response to light illuminating the gemstone. Facets may originate from reflective facets, internal reflective facets, girdle, girdle facets, culet, table, or pavilion. Each type of facet or gemstone feature produces a corresponding light response or results in a pattern when illuminated by incident light.

[0125] Furthermore, combinations of one or more halos, auras, facets, petals, etc., as well as their unions or intersections, can produce a variety of patterns. For example, the crown facets as reflections of an arrow combined with the pavilion facets that contact that reflection can produce an arrow pattern. Alternatively, the bottom view of such a combination can give facet reflections displaying a heart pattern. In one example, a total of eight top views of arrows and eight bottom views of hearts can create a pattern called a heart and arrow, which is caused by the symmetrical cutting and placement of each facet on the crown, pavilion, tabletop, base, and girdle.

[0126] Once the machine learning module 154 is trained to evaluate the patterns of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the pattern of the test gem 150. For example, the output value could be a single number or a letter grade as described above.

[0127] Evaluation size

[0128] The machine learning module 154 can also be trained to objectively measure the size and angles of the test gemstone 150. The size of the gemstone can be its maximum length, width, and height. Angles can be formed between line segments related to the size and characteristics of the test gemstone 150. For example, in a side view, an angle can be formed between the table line and another line segment, forming the crown angle. In another example, the angle can be formed by the pavilion and a line segment from the pavilion to the girdle. Such measured sizes and angles can be used by the machine learning module 154 to determine the volume of the test gemstone 150. In one embodiment, the image capture assembly 120 can control the lens 125 to capture multiple images with different focal length settings. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture assembly 120 to capture multiple training images with different sizes. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0129] Each training image is associated with recognition information representing the size and lines of each different gemstone at each different angle. The recognition information can be at least one numerical value representing the specific size of each different gemstone at each different angle. Alternatively, the recognition information can be a triplet of numerical values ​​representing, in order, the maximum length, maximum width, and maximum height of each different gemstone. Furthermore, the recognition information can be an N-tuple of N numerical values ​​representing, in order, the width of the gemstone's table, the height of the gemstone's crown, the height or thickness of the gemstone's girdle, the height of the gemstone's pavilion, the width of the gemstone's culet, the facet structure of the gemstone, the structure of the gemstone, etc. Additionally, the recognition information can specify the angles between the lines of a given gemstone.

[0130] Once the machine learning module 154 is trained to evaluate the size of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output at least one value that objectively measures the size of the test gem 150. For example, the output value can be a single number or an N-tuple of numbers, as described above. Alternatively, the user can input a value associated with the test gem 150, or a value associated with another gem, against which the test gem 150 can be analyzed after analysis of that other gem.

[0131] Assess symmetry

[0132] The machine learning module 154 can also be trained to objectively measure the symmetry of the test gemstone 150. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different symmetries. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images with different symmetries. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0133] Each training image is associated with recognition information representing the symmetry of each different gemstone at each different angle. The recognition information can be a numerical value representing the symmetry of each different gemstone at each different angle; for example, "1" represents a symmetrical gemstone, and "0" represents a gemstone without symmetry. Furthermore, the recognition information involving symmetry can be a numerical range. For example, this range could be between 0 and 1. Light return or formation can be graded based on the sharpness or intensity of the light return, and the accuracy or lack thereof in the formation and imaging of multiple light returns—such as halos, rings, light petals, etc.—and their patterns and formations. Additionally, facet accuracy information or precision can be input into the dataset to describe how light responds to facets, symmetry, or cutting processes.

[0134] In another example, the numerical value can be greater than 1 or less than 0. Furthermore, negative numbers (-ve numbers) and positive numbers (+ve numbers) can be used, allowing symmetry to be better or worse than a certain range, such as specifying a better gem cut. For example, an older cut can be downgraded, or it can be shown as a positive improvement over a given older symmetry or grade. In additional alternative embodiments, the numerical value can be a decimal, such as 79.95 or 0.7995. Such decimal values ​​can specify many such proportions, combinations, and possibilities of gem length, base, and height. Furthermore, the ratio of length, base, and height can be specified by decimal values. Additionally, the numerical value can specify the table, crown, girdle, pavilion, and facets, as well as percentages, angles, positions, etc. Alternatively, the letter grade can be “T” for “TRUE” indicating a symmetrical gem, and “F” for “FALSE” indicating a non-symmetrical gem. Furthermore, the identification information can be a word or message, such as “TRUE” for a symmetrical gem and “FALSE” for a non-symmetrical gem. Other words that can be used include "ideal," "excellent," "very good," "good," "poor," and "fair." These words or messages can be associated with numerical values ​​to describe quantitative values ​​of symmetry.

[0135] Once the machine learning module 154 is trained to evaluate the light response and pattern symmetry of different gemstones under different angles and lighting conditions using different light and variable focal length enabled, the system 100 can be activated to illuminate the test gemstone 150, such as... Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the symmetry of the test gem 150. For example, the output value could be a single number or a letter grade as described above.

[0136] Assess the cut grade

[0137] The machine learning module 154 can also be trained to objectively measure the cut grade of the test gemstone 150. The cut grade can be nominal. Alternatively, the cut grade can be a final cut grade, which can be a combination of one or all other grades associated with the gemstone. Such other grades can include polish grade, finish grade, symmetry grade, light return grade, size and proportions based on physical calculations or the optical properties of the test gemstone 150. The cut grade can be a weighted average of other grades. Alternatively, the cut grade can be extracted from a user model or a predefined grading model according to a predetermined formula or method. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different cut grades. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images with different cut grades. In one embodiment, the support 102 may be motorized, for example, with a servo motor under the control of the processing unit 124.

[0138] Each training image is associated with recognition information representing the cut grade of each different gemstone at each different angle. The recognition information can be a numerical value representing the cut grade of each different gemstone at each different angle. Alternatively, the recognition information can be a letter grade representing the cut grade of each different gemstone at each different angle.

[0139] Once the machine learning module 154 is trained to evaluate the cut grade of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as... Figure 1A As shown, the image capture component 120 captures a query image of the test gemstone 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the cut grade of the test gemstone 150. For example, the output value can be a single number or a letter grade as described above. The final cut grade can be derived from one or more other grades, such as polish, symmetry, light response, performance, scintillation, brightness, etc.

[0140] Assess clarity

[0141] The machine learning module 154 can also be trained to objectively measure the clarity of the test gemstone 150. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different clarities. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images with different clarities. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0142] Each training image is associated with identification information representing the clarity of each different gemstone at each different angle. The identification information can be a numerical value representing the clarity of each different gemstone at each different angle. This value can be a weighted average of clarity values ​​based on the gemstone's placement, color, size, shape, translucency, opacity, light reflectivity, type, proximity to inclusions, or any other characteristic of the gemstone. Furthermore, the clarity value can be based on the degree of light shining on a particular feature or inclusion. Additionally, the clarity value can be based on the degree to which light bends around a particular feature or inclusion. Furthermore, the clarity value can be based on the degree to which light passes through a particular feature or inclusion. Alternatively, the clarity value can be based on the degree to which light leaks from various parts of the gemstone. Such leaked light can be reflected, refracted, or can illuminate different parts of the gemstone. Furthermore, light can increase or decrease shadows and reflections, or may cause blind spots that can obscure parts of the gemstone, including features or inclusions.

[0143] Clarity values ​​can be affected by light at different distances from features or inclusions, or vary with focal length. Furthermore, clarity values ​​can be influenced by lighting conditions on the gemstone and by prominent features or inclusions. Using such clarity values, 3D models or maps can be created to determine the optimal distance or focal length settings and the best lighting conditions or highlighting methods for measuring gemstone clarity. In an alternative embodiment, clarity values ​​can be based on a user model. The user model can be a predefined model configured to determine clarity values. Alternatively, identification information can be a letter grade representing clarity.

[0144] Once the machine learning module 154 is trained to evaluate the clarity of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as Figure 1AAs shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the clarity of the test gem 150. For example, the output value can be a single number or a letter grade as described above.

[0145] Evaluation of light return

[0146] The machine learning module 154 can also be trained to objectively measure the light return of the test gem 150. Such light return can be associated with one or more patterns, halos, or other lighting effects. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gems to capture multiple training images with different light returns. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gem holder 102 can be configured to move different gems by translation or rotation, allowing the image capture component 120 to capture multiple training images with different light returns. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0147] Each training image is associated with recognition information representing the light return of each different gemstone at each different angle and for each different part of the gemstone. The light return can also be the total light return or partial light return for each different gemstone. The final value of the light return can be a weighted average of the light returns from all parts of the gemstone. Alternatively, the final value can be based on a user-defined formula or a predefined model. The recognition information can be a numerical value representing the light return of each different gemstone or part of the gemstone at each different angle. For example, the numerical value could be "100" representing absolute light return and "0" representing no light return.

[0148] Once the machine learning module 154 is trained to evaluate the light return of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as... Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the light return of the test gem 150. For example, the output value could be a single number as described above.

[0149] Evaluation of modification degree

[0150] The machine learning module 154 can also be trained to objectively measure the refinishing degree of the test gemstone 150. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different refinishing degrees. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images with different refinishing degrees. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0151] Each training image is associated with recognition information representing the degree of refinement of each different gemstone at each different angle. The recognition information can be a numerical value representing the degree of refinement of each different gemstone at each different angle. For example, the value "100" can be assigned to gemstones with absolute refinement, while "0" can be assigned to gemstones with no refinement.

[0152] Once the machine learning module 154 is trained to evaluate the finish of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as... Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the degree of refinement of the test gem 150. For example, the output value could be a single number or a letter grade as described above.

[0153] Assessment and processing

[0154] The machine learning module 154 can also be trained to objectively measure the treatment of the test gemstone 150. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different treatments (e.g., crevice filling, hole filling, laser drilling, gemstone annealing, oiling, or other known treatments). Such treatments can be imaged and taught to the machine learning module 154. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images with different treatments. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0155] Each training image is associated with recognition information representing the treatment of each different gemstone at each different angle. The recognition information can be a numerical value representing the treatment of each different gemstone at each different angle, such as "100" for a clean gemstone and "0" for a completely opaque gemstone. Alternatively, letter grades can be used, such as "A" for a clean gemstone or "F" for a completely opaque gemstone. Furthermore, the letter grades can be "F" for "FALSE" indicating a clean gemstone without treatment, and "T" for "TRUE" indicating an opaque gemstone with treatment.

[0156] Once the machine learning module 154 is trained to evaluate the handling of different gems at different angles, the system 100 can be activated to illuminate the test gem 150, such as Figure 1A As shown, the image capture component 120 captures the query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the performance of the test gem 150. For example, the output value could be a single number or a letter grade as described above.

[0157] Evaluation facets

[0158] The machine learning module 154 can also be trained to objectively measure the facets of the test gemstone 150. In one embodiment, the image capture assembly 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different facets. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture assembly 120 to capture multiple training images with different facets. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0159] Each training image is associated with recognition information representing the facets of each different gemstone at each different angle. The recognition information can be a numerical value representing the facets of each different gemstone at each different angle. For example, the value could be "2" for a gemstone with two facets, "3" for a gemstone with three facets, and so on. Alternatively, the value can be based on the placement and type of facets, as well as the completeness of the facet structure.

[0160] Once the machine learning module 154 is trained to evaluate the facets of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as Figure 1AAs shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the facets of the test gem 150. For example, the output value could be a single number as described above.

[0161] Assessment edge

[0162] The machine learning module 154 can also be trained to objectively measure the edges of the test gemstone 150. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images with different edges. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture component 120 to capture multiple training images with different edges. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0163] Each training image is associated with recognition information representing the edges of each different gemstone at each different angle. The recognition information can be a numerical value representing the edges of each different gemstone at each different angle. For example, the value could be "1" for a gemstone with one edge, "2" for a gemstone with two edges, "3" for a gemstone with three edges, and so on. Alternatively, the value can specify the same type of edge repeated N times.

[0164] Once the machine learning module 154 is trained to evaluate the edges of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the edges of the test gem 150. For example, the output value could be a single number as described above.

[0165] Assess carat weight

[0166] The machine learning module 154 can also be trained to objectively measure the carat weight or mass of the test gemstone 150. In one embodiment, the support 102 may include a weighing device, such as a balance, for measuring the carat weight of the test gemstone 150. The system 100 then captures multiple training values ​​as the weights or mass of different gemstones.

[0167] In another embodiment, carat weight can be determined from the volume or specific gravity of the test gemstone. Furthermore, the machine learning module 154 can be trained to objectively determine the type of the test gemstone. At least one image needs to be obtained, such as a top view, bottom view, side view, and views from different angles. The collected information may include various angles of the test gemstone 150, the length, base, and height of the test gemstone 150, and the facets of the test gemstone 150. Using the collected information, the machine learning module 154 can calculate the volume, size, angles, and weight of the test gemstone 150 from multiple images. Therefore, the image capture assembly 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones to capture multiple training images representing different carat weights with different volumes. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture assembly 120 to capture multiple training images representing different carat weights with different volumes. Images can be taken of objects of known size and dimensions. Alternatively, the user can input the known size or dimensions of the object. Using the input size and dimensions, system 100 can extract the actual size and dimensions using a predefined model or formula, from which the actual weight of the gemstone can be determined. In one embodiment, the support 102 can be motorized, for example, equipped with a servo motor under the control of processing unit 124.

[0168] Each training image is associated with recognition information representing the carat weight of each different gemstone at each different angle. The recognition information can be a numerical value representing the carat weight of each different gemstone. For example, the value could be "2" for a two-carat gemstone and "14" for a fourteen-carat gemstone.

[0169] Once the machine learning module 154 is trained to evaluate the carat weight of different gemstones with different masses or volumes at different angles, the system 100 can be activated to weigh or illuminate the test gemstone 150, such as Figure 1A As shown, the image capture component 120 captures a query image of the test gemstone 150. The trained machine learning module 154 then processes the measured weight or query image to obtain and output a value that objectively measures the carat weight of the test gemstone 150. For example, the output value could be a single number as described above.

[0170] Distinguishing between natural and synthetic gemstones

[0171] The machine learning module 154 can also be trained to objectively evaluate whether the test gemstone 150 is a natural or synthetic gemstone, such as a gemstone manufactured using CVD, HPHT, or other known manufacturing techniques. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to guide light at different angles and with different degrees of polarization, or with zero polarization, in the form of halos, rings, etc., to different gemstones to capture multiple training images with various parameters—e.g., scintillation, color, pattern, light return pattern in the form of halos, rings, etc., size, symmetry, cut grade, clarity, light return, light return pattern in the form of halos, rings, etc., finish, treatment, facets, edges, carat weight, etc. In one embodiment, such as Figure 2D As shown, images 230-1 to 230-3 of two round diamonds were captured under different degrees of polarization. Image 230-1 shows a natural diamond on the left and a CVD lab-grown diamond on the right, both in focus. Image 230-2 shows images of these diamonds in an unpolarized, defocused state, while image 230-3 shows images of these diamonds in a polarized, defocused state. The halos, auras, etc., and pixelation formed by light emitted from the gemstones can be extracted as light patterns and light colors, allowing a neural network 156 to be trained to distinguish between natural and synthetic gemstones. The color of the light, such as a blue or gray hue, depends on the type or treatment of the lab-grown gemstone. A halo of hue will appear in the image or pattern.

[0172] In another embodiment, such as Figure 2E As shown, images of gemstones can be obtained at different focal length settings to distinguish between natural and synthetic gemstones. Images 240-1 to 240-3 show princess-cut diamonds. Image 240-1 is a focused image of a synthetic CVD diamond, while image 240-2 is a defocused image of this synthetic square princess CVD diamond. In the defocused image 240-2, a square black center is visible, creating light reflection formation under defocus imaging. Image 240-3 is a defocused image of three princess-cut diamonds. The bottom defocused diamond has a square black center, while the top two defocused diamonds do not, indicating that the top two diamonds are natural diamonds. Using such defocused images 240-2 and 240-3 with associated identification information for each diamond—i.e., natural or synthetic—a neural network 156 can be trained to identify and distinguish between natural and synthetic gemstones. Figure 2EIn the image examples, the trained neural network 156 can be trained to recognize the presence of such square and black central regions in images 240-2 and 240-3. Therefore, the trained neural network 156 can determine that those portions of images 240-2 and 240-3 with square and black centers are associated with synthetic diamonds, while those portions of image 240-3 lacking such square and black centers are associated with natural diamonds. Other patterns can also be found in the images, such as the position and pattern of halos and auras, the formation of halos and auras, the formation patterns of halos and auras, the absorption of ambient light and colors, pixelation, the colors of such halos and pixelation, and patterns of light reflection. In particular, the patterns formed by synthetic gemstones differ and are more diverse than those formed by natural gemstones, and can be recognized by the trained neural network 156.

[0173] Lens 125 or light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using processing unit 124. Alternatively, gem holder 102 can be configured to move different gems by translation or rotation, allowing image capture assembly 120 to capture multiple training images with various parameters. In one embodiment, holder 102 can be motorized, for example, with a servo motor under the control of processing unit 124.

[0174] Each training image is associated with recognition information representing different parameters of each different gemstone at each different angle. The recognition information can be a numerical value representing whether each different gemstone with various parameters is a natural or synthetic gemstone. For example, a value of "1" could represent a natural gemstone and "0" a synthetic gemstone. Additionally, a value of "2" could indicate a high probability but requiring further examination. Alternatively, the recognition information can be a letter grade, such as "A" for a natural gemstone and "B" for a synthetic gemstone. Furthermore, the recognition information could be "T" for "TRUE" indicating a natural gemstone and "F" for "FALSE" indicating a synthetic gemstone. Additionally, the recognition information can be a word or message, such as "TRUE" for a natural gemstone and "FALSE" for a synthetic gemstone. The recognition information can also include the word "SUSPECT," indicating that the probability of the tested gemstone 150 being a natural gemstone is high, but the definitive conclusion is uncertain.

[0175] Once the machine learning module 154 is trained to evaluate various parameters of different gems at different angles, the system 100 can be activated to illuminate the test gem 150, such as Figure 1AAs shown, the image capture component 120 captures a query image of the test gemstone 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures various parameters of the test gemstone 150 and indicates whether the test gemstone 150 is a natural or synthetic gemstone. For example, the output value can be a single number, a letter grade, or an indicative word such as “TRUE” or “FALSE”, as described above.

[0176] Evaluation of imaging parameters

[0177] The machine learning module 154 can also be trained to objectively measure the imaging parameters of the system 100 when evaluating the test gem 150. For example, once trained, the machine learning module 154 can objectively measure light parameters, objectively measure camera parameters (e.g., the focus of the image capture component 120), or the focus level of any identified parameters, objects, or information, or objectively measure the relative positions of the light source 160, the image capture component 120, and the test gem 150. For example, the gem may be in focus overall, but a facet, a portion of the gem, or an inclusion may be out of focus, or vice versa. The machine learning module 154 can be trained to understand the degree to which the gem or a portion thereof is in focus or out of focus. In one embodiment, the image capture component 120 can control the lens 125 or the light source 160 to direct light at different angles to different gems to capture multiple training images with different imaging parameters. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gem holder 102 can be configured to move different gems by translation or rotation, allowing the image capture assembly 120 to capture multiple training images with different imaging parameters. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124.

[0178] Each training image is associated with recognition information, which represents the imaging parameters of each different gemstone at each different angle. The recognition information can be a numerical value representing the imaging parameters of each different gemstone at each different angle. For example, camera parameters could be a numerical measure of the focal length setting of the image capture component.

[0179] Once the machine learning module 154 is trained to evaluate the imaging parameters of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150, such as... Figure 1A As shown, the image capture component 120 captures a query image of the test gem 150. The trained machine learning module 154 then processes the query image to obtain and output a value that objectively measures the imaging parameters of the system 100 when evaluating the test gem 150. For example, the output value could be a single number or a letter grade as described above.

[0180] Distinguish the gemstone from its setting and other gemstones.

[0181] The machine learning module 154 can also be trained to objectively distinguish between the test gemstone 150 and the gemstone setting or other gemstones. Figure 1A The support 102 in the gemstone can be one or more prongs, such as metal prongs that hold the test gemstone 150 in jewelry (like a ring). For example, once training is complete, the machine learning module 154 can objectively detect the test gemstone 150, rather than the gemstone support or other gemstones. Examples of systems and methods for capturing and analyzing single or batch processing of gemstones or jewelry articles using machine vision and other technologies are shown and described in commonly pending and commonly assigned U.S. Patent No. 11,222,420 (titled “SYSTEM AND METHOD FOR PROCESSING MULTIPLE LOOSE GEMSTONES USINGIMAGE-BASED ANALYSIS TECHNIQUES”, issued January 11, 2022) and U.S. Patent No. 11,132,779 (titled “JEWELRY ITEMGRADING SYSTEM AND METHOD”, issued September 28, 2021), the entire contents of which are incorporated herein by reference.

[0182] In one embodiment, the image capture assembly 120 can control the lens 125 or the light source 160 to direct light at different angles to different gemstones, gemstone holders, or objects in an image to capture multiple training images with various parameters, such as scintillation, color, pattern, various patterns formed by reflected light and captured by the imaging system, size, physical properties, symmetry, cut grade, clarity, light return, finish, treatment, facets, edges, carat weight, etc., as well as other physical and light response properties of various objects in the images. The lens 125 or the light source 160 can be motorized, for example, with a servo motor. Such motorization can be controlled using the processing unit 124. Alternatively, the gemstone holder 102 can be configured to move different gemstones by translation or rotation, allowing the image capture assembly 120 to capture multiple training images with various parameters. In one embodiment, the holder 102 can be motorized, for example, with a servo motor under the control of the processing unit 124. In another embodiment, the gemstones can be manually fixed or manipulated to create training images. In a further embodiment, the dataset of training images can be obtained from a user's phone—such as a smartphone, a computer connected to a network, or other data collection device.

[0183] Each training image is associated with recognition information representing various parameters of each different gemstone at each different angle. Recognition information can be a numerical value representing various parameters of each different gemstone, each gemstone setting, and each type of gemstone setting at each different angle. Such gemstone setting parameters can indicate a single prong, a group of prongs, a single setting, or a group of settings. A group of settings can specify a single setting or a batch of gemstone settings. A batch of gemstone settings can be identified by bounding boxes or edge drawing. Multiple different halos, or the union of the intersections of rays returned by halos / shapes / patterns, or their differences, can be identified and distinguished. Different objects have different types of light returns, halos, and patterns, which are identified and labeled in the training dataset. For example, setting 102 can be assigned the value "0", gemstone 150 can be assigned the value "1", and at least the second gemstone can be assigned the value "2". Other values, such as "3", "4", etc., can be assigned to other elements associated with the gemstone, such as prongs or metal parts used for setting the gemstone.

[0184] Once the machine learning module 154 is trained to evaluate various parameters and light responses of different gemstones at different angles, the system 100 can be activated to illuminate the test gemstone 150 or the support 102, such as Figure 1A As shown, or at least a second gem. Image capture component 120 captures a query image of test gem 150, support 102, or at least a second gem. The query image may capture one or more test gems, gem supports, and test objects. Then, trained machine learning module 154 processes the query image to obtain and output a value that objectively measures the presence or absence of gem 150, prongs, support 102, or at least a second gem, and generates an output value. Alternatively, trained machine learning module 154 may objectively measure the absence of any gem, prong, or gem support. For example, the output value may be a single number as described above.

[0185] Gemstone Safety Analysis

[0186] The system 100, equipped with a post-trained machine learning module 154, can objectively measure various parameters of the test gemstone 150, as described above, including but not limited to scintillation, color, pattern, size, symmetry, cut grade, clarity, light return, finish, treatment, facets, rim, carat weight, imaging parameters, and whether the test gemstone 150 is a natural or synthetic gemstone. Each parameter of the test gemstone 150 is uniquely associated with the gemstone 150 and its image. Therefore, various parameters, light response, pattern, halo, and other formations can be captured and analyzed by the system 100, which can act as a fingerprint to uniquely identify one gemstone from another. Thus, secure analysis of the test gemstone 150 can be performed based on the image of the gemstone 150 processed by the system 100. Alternatively, such images can also be received and processed by other systems outside the system 100. External systems may include external databases, or a user's telephone, smartphone, or image capture device configured to receive and analyze images. Furthermore, the system 100 can maintain an image database or receive images from an external database. Alternatively, a blockchain can be maintained for such images to securely store such images and objectively measured parameters for secure identification of the test gem 150 in the future. Using such images stored in a database or blockchain, system 100 or an external system can verify one or more received test images based on the images in the database or blockchain.

[0187] A blockchain is a distributed ledger composed of a constantly growing list of records—called blocks—securely linked together using cryptography. For example, a blockchain containing an image of a given gemstone and its associated objective measurement parameters could be stored... Figure 1B The storage is located in memory 122 or external storage 116. A computing device, such as a smartphone, tablet, or personal computer, may have the ability to access the blockchain via network 114. This computing device may be part of an external gemstone evaluation platform 118. Thus, for example, a gemstone appraiser can use such a computing device to verify a given gemstone by accessing a secure blockchain trace, based on unique and objectively measured parameters evaluated by a trained machine learning module 154.

[0188] Any determination of subjective or objective results regarding a gemstone can be transmitted, examined, or verified using a smartphone app, a computer connected to a network, or other known information transmission devices and systems. Any other user can access such results to verify, identify, match, measure, value, sell, or grade the evaluated gemstone, or to perform any other alternative task. Results can be accessed via the internet, other networks, or other platforms that facilitate information exchange.

[0189] It is understood that the foregoing embodiments provide a solution, among others, for objectively classifying and quantifying gemstone grading and the 4Cs of gemstone aesthetics and other natural characteristics and qualities using optical responses in gemstone images captured under various lighting, imaging systems, and methods. Other disclosed methodologies include advanced gemstone counting, gemstone identification, gemstone security analysis, and techniques for distinguishing between natural, lab-grown, synthetic, imitation, or treated gemstones.

[0190] It should be noted that although the foregoing description pertains to a gem imaging device and a method for gem analysis using that device, the systems and methods disclosed herein can be similarly deployed or implemented in situations, contexts, and settings far beyond the reference scenario. It should be understood that the same numbers throughout the figures represent the same elements, and not all components and / or steps described and illustrated in all the figures are necessary for all embodiments or arrangements.

[0191] Therefore, the illustrative embodiments and arrangements of this system and method provide systems, processes, and computer-implemented control methods, computer systems, and computer program products for gemstone processing. In alternative embodiments, the system and method can be applied to process non-gemstone objects, or combinations of gemstone and non-gemstone objects. The flowcharts and block diagrams in the various figures illustrate the architecture, functionality, and operation of possible implementations of the system, method, and computer program product according to various embodiments and arrangements. In this regard, with respect to computer-implemented methods, each block in a flowchart or block diagram may represent a module, segment, or code portion that includes one or more executable instructions for implementing a specified logical function.

[0192] It should also be noted that in some alternative implementations, the functions described or illustrated in the block diagram may occur in a different order than described. For example, two blocks or operations that are shown or described consecutively may actually be executed substantially concurrently, or sometimes in reverse order, depending on the functions involved. It should also be noted that, where applicable, function blocks or operations may be implemented by a dedicated hardware-based system, or a combination of dedicated hardware and computer instructions, that performs the specified functions or actions.

[0193] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the terms “comprising” and / or “including” as used in this specification specify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0194] Furthermore, the wording and terminology used herein are for descriptive purposes and should not be considered limiting. The terms “including,” “comprising,” “having,” “containing,” “involving,” and variations thereof, as used herein, are intended to cover the items listed thereafter and their equivalents, as well as additional items.

[0195] The foregoing subject matter is provided by way of illustration only and should not be construed as limiting. Various modifications and changes may be made to the subject matter described herein without following exemplary embodiments and applying the described content, and without departing from the true spirit and scope of the invention as defined by the statements in the following group of claims and equivalent structures, functions or steps as covered by this disclosure.

Claims

1. A method for evaluating a gemstone from gemstone images, comprising: receiving, at a processor of a computing device, a query image set comprising at least one query image of a gemstone, the computing device having a non-transitory computer readable storage medium and the processor configured by executing a software program stored in the storage medium; analyzing the query image set using a trained machine learning algorithm, wherein the machine learning algorithm is trained to detect at least one gemstone feature depicted within a query image and is trained based on a plurality of training image sets for gemstones, each training image set being captured using an image capture device from a respective gemstone and each training image set comprising a plurality of images of the respective gemstone captured at different focal length settings; identifying, by the trained machine learning algorithm, one or more of the at least one gemstone feature in the query image set comprising the at least one query image; and outputting a notification of the at least one gemstone feature identified.

2. The method of claim 1, further comprising: receiving, with the processor, the training image sets; and training, with the processor, the machine learning algorithm using the received training image sets, the machine learning algorithm trained to detect the at least one gemstone feature.

3. The method of claim 1, further comprising: capturing, with the processor, the query image set comprising the at least one query image using an image capture device. A given training image set comprising images captured at different focal length settings comprises: at least one in-focus image, at least one under-focused image, and at least one over-focused image.

4. The method of claim 1, wherein, The at least one gemstone feature is selected from the group consisting of: an inclusion of the gemstone, a particle on the gemstone, a polish mark of the gemstone, a scratch on the gemstone, an internal pattern of the gemstone, a color of an inclusion of the gemstone, a clarity of the gemstone, a fire of the gemstone, a brilliance of the gemstone, a sparkle of the gemstone, a scintillation of the gemstone, a color of the gemstone, a cut of the gemstone, a symmetry of the gemstone, a polish of the gemstone, a facet of the gemstone, an edge of the gemstone, a shape of the gemstone, a halo of the gemstone, a pattern of the gemstone, and a color change of the gemstone.

5. The method of claim 2, wherein, 6. The method of claim 2, further comprising: providing, to the machine learning algorithm, ground truth information for each of the training image sets, the ground truth information identifying one or more of the at least one gemstone feature of the respective gemstone corresponding to each training image. The ground truth information further comprises a respective focal length setting corresponding to each training image.

7. The method of claim 6, wherein, The ground truth information for the respective gemstone corresponding to a given training image comprises:

8. The method of claim 6, wherein, a description of a high-level feature of the respective gemstone selected from the group consisting of: clarity, fire, color, brilliance, scintillation, shape, and cut; a location and classification of one or more of the following features of the respective gemstone within the given training image: ​ a color of an inclusion of the respective gemstone, an inclusion of the respective gemstone, a facet of the respective gemstone, an edge of the respective gemstone, a shape of the respective gemstone, and a color variation of the respective gemstone; and a location of one or more features depicted within the given training image caused by a reflection, refraction, diffraction, or transmission of light by the respective gemstone.

9. The method of claim 1, wherein, the machine learning algorithm is selected from the group consisting of a convolutional neural network, a deep neural network, an artificial immune system (AIS), a you only look once (YOLO) module, a neural Turing machine (NTM), a differential neural computer (DNC), a support vector machine (SVM), a deep learning neural network (DLNN), a naive Bayes module, a decision tree module, a logic model tree induction (LMT) module, an NBTree classifier, a case-based module, a linear regression module, a Q-learning module, a temporal difference (TD) module, a deep adversarial network, a fuzzy logic module, a K- nearest neighbor module, a clustering module, a random forest module, and a rough set module.

10. The method of claim 3, further comprising: illuminating the query gemstone using a light source selected from the group consisting of an incandescent lamp, a light emitting diode, and a laser.

11. The method of claim 1, wherein, the set of query images includes a plurality of query images captured at different focal length settings, and wherein the step of identifying one or more of the at least one gemstone feature is performed using the set of query images.

12. The method of claim 1, wherein, a given set of training images includes images captured at different lighting conditions.

13. A system for evaluating a gemstone from a gemstone image, comprising: an image capture device having a plurality of different focal length settings and configured to capture the gemstone image of a gemstone; and a gemstone evaluation device comprising: a processing unit including a machine learning algorithm, wherein the machine learning algorithm is trained based on a set of training images for a plurality of gemstones, each set of training images being captured from a respective gemstone using the image capture device, each set of training images including a plurality of images of the respective gemstone captured at different focal length settings, and wherein the machine learning algorithm is trained to detect at least one gemstone feature depicted within a query image, and wherein the processing unit is configured to: receive, from the image capture device, a set of query images including at least one query image of a gemstone, and analyze the set of query images using the trained machine learning algorithm, wherein the trained machine learning algorithm is configured to identify, based on the query image, one or more of the at least one gemstone feature in the set of query images including the at least one query image; and an output device configured to output a notification of the identified at least one gemstone feature.

14. The system of claim 13, wherein, A given training image set comprising images captured at different focal settings includes at least one in-focus image, at least one under-focused image, and at least one over-focused image.

15. The system of claim 13, wherein, The machine learning algorithm is trained from the training image set and, for each of the training image set, ground truth information identifying one or more gem features of the at least one gem feature of the respective gem corresponding to each training image.

16. The system of claim 15, wherein, The ground truth information further includes a respective focal setting corresponding to each training image.

17. The system of claim 15, wherein, The ground truth information for the respective gem corresponding to a given training image includes: a description of a high-level feature of the respective gem selected from the group consisting of: clarity, fire, color, and cut; a location and classification of one or more of the following features of the respective gem within the given training image: a grain on the gem, a polish mark of the gem, a scratch of the gem, an internal pattern of the gem, a color of an inclusion of the gem, an inclusion of the gem, a facet of the gem, a girdle of the gem, a shape of the gem, and a color change of the gem; and a location of one or more reflections caused by a reflection, refraction, diffraction, or transmission of light by the gem depicted within the given training image.

18. The system of claim 13, further comprising: a light source configured to illuminate the gem, the light source selected from the group consisting of: an incandescent lamp, a light emitting diode, and a laser.

19. The system of claim 13, wherein the at least one gem feature is selected from the group consisting of: an inclusion of the gem, a grain on the gem, a polish mark of the gem, a scratch of the gem, an internal pattern of the gem, a color of an inclusion of the gem, a clarity of the gem, a fire of the gem, a brilliance of the gem, a sparkle of the gem, a fire of the gem, a color of the gem, a cut of the gem, a symmetry of the gem, a facet of the gem, a girdle of the gem, a shape of the gem, and a color change of the gem.

20. The system of claim 13, wherein, the machine learning algorithm is selected from the group consisting of: a convolutional neural network, a deep neural network, an artificial immune system (AIS), a you only look once (YOLO) module, a neural Turing machine (NTM), a differential neural computer (DNC), a support vector machine (SVM), a deep learning neural network (DLNN), a naive Bayes module, a decision tree module, a logistic model tree induction (LMT) module, an NBTree classifier, a case-based module, a linear regression module, a Q-learning module, a temporal difference (TD) module, a deep adversarial network, a fuzzy logic module, a K-nearest neighbor module, a clustering module, a random forest module, and a rough set module.

21. The system of claim 13, wherein, the query image set includes a plurality of query images captured at different focal settings, and wherein the step of identifying one or more gem features of the at least one gem feature is performed using the query image set.

22. The system of claim 13, wherein, A given training image set includes images captured at different lighting conditions or different focal conditions.

Images