Systems and methods for gemstone identification

Abstract

The present invention provides a computer-implemented method for identifying a specific gemstone having known reference points. In one embodiment, the computer-implemented method comprises the following steps: (a) detecting the known reference points from a source image of the specific gemstone; (b) retrieving one or more reference images of possible gemstones from a gemstone database based on the known reference points; (c) extracting edge features from the source image and the one or more reference images; (d) calculating a similarity value by pairing the edge features of the one or more reference images with those of the source image; and (e) identifying the specific gemstone from the one or more reference images based on the similarity value.

Description

Technical Field

[0001] The present invention relates to systems and methods for gemstone identification. Background Art

[0002] Gemstones, such as diamonds, are highly valued items sought after by collectors and investors due to their rarity, beauty, and durability. However, the high value of gemstones makes them a lucrative target for counterfeiters and fraudsters, who substitute counterfeit gemstones for genuine gemstones during storage, transportation to stores, and display in stores.

[0003] Currently, gemstone identification typically involves complex gemstone examinations by trained gemologists using specialized equipment such as microscopes. However, this process can be time-consuming, expensive, and difficult to implement, particularly in a store setting. Therefore, there is a need for a portable, efficient gemstone identification method that can be easily performed by non-experts and applied in stores without a constant internet connection.

[0004] Edge devices, as defined in this disclosure, are devices capable of local data processing and cloud connectivity. Edge devices are designed to improve efficiency and provide real-time processing capabilities, while maintaining high performance even when connectivity to the cloud is limited. The gem identification system's edge capabilities enable local processing of streaming video data, reducing the need for high bandwidth and cloud computing power. This makes the system ideal for use in stores and other locations where a continuous cloud connection may not be available.

[0005] The physical characteristics of a cut gemstone are a unique combination that can be used to identify that gemstone. Traditional methods include examination by a gemologist to detect variations in the gemstone’s size, table facets, angles, inclusions, and other physical characteristics.

[0006] Some gemstones also carry their certification mark. For example, diamonds certified by reputable laboratories like GIA and IGI have their certification number inscribed on the girdle. These markings are applied via laser inscription, most commonly on the girdle. Laser inscription is a permanent gemstone certification, provided the diamond has not been processed or re-polished.

[0007] Recent advances in computer vision and machine learning technologies have made it possible to authenticate gemstones using image matching, based on each gemstone's unique physical characteristics and markings. Image matching involves comparing the unique features of a gemstone captured in an image with a database of known authentic gemstones to determine its authenticity. Image matching can also be extended to matching streaming video to images, as routinely demonstrated by facial recognition systems.

[0008] U.S. Patent Publication No. 11,037,282B2 demonstrates the use of machine learning algorithms for the identification of clarity features and inclusions. Of particular note is the machine learning algorithm's ability to extract useful information related to edges. The present invention leverages this machine learning algorithm's ability to extract edges as features, and its robustness to data variations without degradation, and applies it to gem identification.

[0009] U.S. Patent Publication No. 11,232,553 B2 further discloses a jewelry security analysis system, method, and computer program product that allows a user to authenticate jewelry by scanning at least the girdle of a gemstone. This invention expands on the concept of capturing a portion of a gemstone to identify unique characteristics, enabling gemstone authentication by capturing only unique markings, such as a certificate number and brand label, along with the surrounding area. Summary of the Invention

[0010] The present invention provides a computer-implemented method for identifying a specific gemstone having known reference points. In one embodiment, the computer-implemented method comprises the following steps: (a) detecting the known reference points from a source image of the specific gemstone; (b) retrieving one or more reference images of possible gemstones from a gemstone database based on the known reference points; (c) extracting edge features from the source image and the one or more reference images; (d) calculating a similarity value by pairing the edge features of the source image and each of the one or more reference images; and (e) identifying the specific gemstone from the one or more reference images based on the similarity values.

[0011] The present invention also provides a non-transitory computer-readable storage medium having stored thereon instructions, which, when executed by one or more processors, causes the one or more processors to perform operations including the computer-implemented method according to the present invention.

[0012] The present invention further provides a system for identifying a specific gemstone, the system using the computer-implemented method of the present invention. In one embodiment, the system includes: (a) one or more devices for acquiring a source image, each of the one or more devices including: (i) a first processor; (ii) a memory unit having executable instructions stored therein, the instructions causing the first processor to perform step (a) of the computer-implemented method when executed; (iii) a first communication module; (b) a server including a second processor, the gemstone database, a second communication module, and a memory unit having executable instructions stored therein, the executable instructions causing the second processor to perform steps (b) to (e) of the computer-implemented method when executed; the first communication module transmitting the detected known reference point to the second communication module; and the second communication module transmitting the identified specific gemstone to the first communication module. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention is described in the form of annotated system schematics and module drawings. Further embodiments are described through user operation flow charts and user interface wireframe diagrams.

[0014] Figure 1 This is a flow chart of the identification method. This procedure will begin when the user enters the identification process.

[0015] Figure 2 It is a user operation flow chart describing the operation procedure of the present invention.

[0016] Figure 3 The invention relates to the electrical connection between the modules in the hardware housing of the system and the rechargeable battery device.

[0017] Figure 4 This image shows a successful match between a gemstone sample and a database reference image, even though dust particles are present in the test image. This image also illustrates the AI edge detection algorithm's identification process and graphical results, as well as the AI matching algorithm's filtering results.

[0018] Figure 5 This figure shows the identification results of gemstone samples that do not match the reference image records in the database. This figure also further illustrates the identification process and graphical results of the AI edge detection algorithm, as well as the filtering results of the AI matching algorithm.

[0019] Figure 6 The table shows the experimental results of testing an experimental dataset using the method of the present invention, which consists of paired detection images of real diamond samples and reference images, and further compares the accuracy obtained by the method of the present invention with that of a method lacking the key step.

[0020] Figure 7 The table shows the experimental results of the method of the present invention on a dataset of real diamond sample reference images. The method of the present invention successfully matched the test image with the correct reference image without any false positive results.

[0021] Figure 8 The table below demonstrates the method's ability to handle various images and lighting conditions. The table shows the method's performance under specified brightness and contrast conditions after varying the original image by a certain percentage. It also displays the resulting detection image using these specified brightness and contrast conditions. DETAILED DESCRIPTION

[0022] The present invention relates to a gemstone identification system and method utilizing video-to-image matching. The system comprises a video recording device equipped with an adjustable work surface for viewing gemstones at various angles, a wireless transceiver for internet and intranet connectivity, and the necessary computing equipment for running artificial intelligence algorithms. The gemstone identification method of the present invention involves detecting and identifying gemstones and their corresponding unique identifying marks or features, or both, extracting edge features, matching these edge features with reference images, and outputting a similarity score to determine a pass or fail.

[0023] The present invention provides a computer-implemented method for identifying a specific gemstone having a known reference point. In one embodiment, the computer-implemented method comprises the following steps: (a) detecting the known reference point from a source image of the specific gemstone; (b) retrieving one or more reference images of possible gemstones from a gemstone database based on the known reference point; (c) extracting edge features from the source image and the one or more reference images; (d) calculating a similarity value by pairing the edge features of the source image and each of the one or more reference images; and (e) identifying the specific gemstone from the one or more reference images based on the similarity value.

[0024] In one embodiment, the known reference point is an inscribed mark or a natural mark.

[0025] In one embodiment, the source image is a still image from a camera or an image obtained from streaming video processing.

[0026] In one embodiment, one or more of steps (a), (b), (c) or (d) are performed using an artificial intelligence algorithm.

[0027] In one embodiment, the artificial intelligence algorithm is trained with a plurality of labeled images of landmarks and gemstones along with data augmentation.

[0028] In one embodiment, one or more of steps (a), (b), (c) or (d) includes filtering image noise.

[0029] In one embodiment, the known reference point is located on, near, or outside the girdle of the particular gemstone.

[0030] In one embodiment, step (b) comprises generating a string of machine-readable text based on the record of the known reference points, and retrieving the one or more reference images recorded together with the string of machine-readable text.

[0031] In one embodiment, step (d) comprises aligning edge features of the source image and the one or more reference images. In another embodiment, step (d) further comprises using edge features from the one or more reference images as a hard filter on the source image.

[0032] In one embodiment, step (e) comprises retaining only images whose similarity value exceeds a threshold.

[0033] The present invention also provides a non-transitory computer-readable storage medium including stored instructions, characterized in that when executed by one or more processors, the one or more processors are caused to perform operations including the computer-implemented method according to claim 1.

[0034] The present invention further provides a system for identifying a specific gemstone using the computer-implemented method of the present invention. In one embodiment, the system comprises: (a) one or more devices for acquiring a source image, each of the one or more devices comprising: (i) a first processor; (ii) a memory unit having executable instructions stored therein, wherein the instructions, when executed, cause the first processor to perform step (a) of the computer-implemented method; (iii) a first communication module; (b) a server comprising a second processor, the gemstone database, a second communication module, and a memory unit having executable instructions stored therein, wherein the executable instructions, when executed, cause the second processor to perform steps (b) to (e) of the computer-implemented method; the first communication module transmitting the detected known reference point to the second communication module; and the second communication module transmitting the identified specific gemstone to the first communication module.

[0035] In one embodiment, the first communication module or the second communication module is a wireless communication module.

[0036] In one embodiment, each of the one or more devices comprises a camera or a video recorder for obtaining the source image.

[0037] In one embodiment, each of the one or more devices comprises lighting for providing optimal lighting conditions for capturing the source image.

[0038] In one embodiment, the system further comprises an adjustable stage for positioning the particular gemstone to obtain the source image.

[0039] In one embodiment, at least one of the one or more devices is a portable device.

[0040] In one embodiment, the server is a cloud server.

[0041] The present invention also provides an automatic moving object recognition system and method for gemstone identification. In one embodiment, the gemstone identification method includes: (a) using a video camera to capture a streaming video of a loose stone or gemstone on a piece of jewelry being inspected; (b) using the captured image of each unique gemstone as a reference image for identification; (c) using an artificial intelligence detection and recognition algorithm to identify unique identifying marks and features of the loose stone or gemstone on the inspected loose stone or jewelry and converting them into machine-readable text, which is used to obtain a set of reference images of the corresponding gemstone; (d) using an artificial intelligence algorithm to identify, recognize, and extract edge features, and to instantly match the physical features of the inspected gemstone depicted in the inspection image with the identified corresponding reference image set, with the matching results being scored based on the dissimilarity between the inspected gemstone and the given image set; and (e) using a threshold value of the matching score to determine whether the identification is passed or failed.

[0042] In one embodiment, the gemstone can be any type of gemstone, any type of cut, attached to another object, or loose. The gemstone must contain at least one unique identifying mark or feature, or both, for identification.

[0043] In one embodiment, the streaming video includes digital video.

[0044] In one embodiment, the reference image consists of a digital image.

[0045] In one embodiment, each set of reference images requires only one image of the gemstone's unique identifying mark or feature, or both, and the immediate surrounding area. More than one image of the object being inspected may be used, but this is not strictly required by the method.

[0046] In one embodiment, the artificial intelligence detection and identification algorithm is trained using images of labeled gemstones and unique identifying marks or features, or both, to detect and identify gemstones, unique identifying marks or features, or both through data augmentation.

[0047] In one embodiment, the training method of the present invention is used, wherein the artificial intelligence detection and recognition algorithm is flexible in terms of the gemstone's camera angle, distortion and lighting conditions, image quality, and noise when recognizing and identifying the unique identification marks and characteristics of the gemstone.

[0048] In one embodiment, the artificial intelligence detection and recognition algorithm includes steps to identify key landmarks and perform image corrections that improve the matching capabilities of the present invention.

[0049] In one embodiment, the present invention uses an artificial intelligence edge detection algorithm to extract edge features from a detected image and a set of reference images. The artificial intelligence edge detection algorithm is trained using a set of reference images and their corresponding annotated edges.

[0050] In one embodiment, an artificial intelligence edge detection algorithm using the training method of the present invention has a certain degree of flexibility in extracting edge features in terms of the shooting angle, distortion, lighting conditions, quality, and noise of the input image without performance degradation.

[0051] In one embodiment, the present invention uses an artificial intelligence matching algorithm to calculate the similarity score, wherein the artificial intelligence matching algorithm is trained using edge features of the original image and the data-augmented reference image.

[0052] In one embodiment, the method includes the step of filtering out noisy edge features in the gemstone image by hard filtering using edge features from a reference image.

[0053] In one embodiment, the method includes the step of filtering out noisy edge features, which improves the matching capability of the present invention.

[0054] In one embodiment, the artificial intelligence matching algorithm of the method of the present invention has a certain degree of flexibility to handle noise in the extracted features without degrading the matching performance.

[0055] In one embodiment, the preset similarity threshold is based on domain knowledge and security requirements of the entity operating the system.

[0056] In one embodiment, the matching score is calculated by calculating the similarity between the inspected gemstone and the reference image set, where a matching score of 0 represents no match and a matching score of 1 represents a match.

[0057] The present invention also provides a system for gem identification, the system comprising but not limited to: (a) a high-resolution video recording device for recording a streaming video of the physical characteristics and certificate number of the gemstone. The gemstone can be of any type described in the present invention. (2) a processing unit including a microprocessor and memory capable of running a computer program for artificial intelligence detection and identification algorithms. (3) a wireless transceiver for Internet and intranet connection. (4) a battery device for providing operating power to the system. (5) a display, the display being used to: (i) provide a medium for user control of the system and the associated computer program. (ii) display the streaming video captured by the video recording device during the gem inspection mode and certification identification process.

[0058] In one embodiment, the system's high-resolution digital video camera can record a streaming video of the gemstone and its laser-inscribed certificate number. The distance between the video camera and the gemstone can be adjusted by a knob.

[0059] In one embodiment, the adjustable work surface can be adjusted to different heights and orientations through physical interaction by the user, providing different perspectives of the identified loose stone or gemstone in jewelry.

[0060] In one embodiment, the processing unit of the system can execute a computer program that runs the necessary artificial intelligence algorithms, streaming video processing programs, and video to image matching programs on the system.

[0061] In one embodiment, the wireless transceiver module of the system can perform wireless transmission of streaming video, wireless transmission and reception of software updates, wireless transmission and reception of user information, and wireless reception of authentication results.

[0062] In one embodiment, the cloud database stores one or more gemstone images, wherein at least one image records the certificate number of the gemstone. In another embodiment, the database stores the corresponding certificate number as a string for each set of images to record the gemstone.

[0063] In one embodiment, the battery system is rechargeable and can provide sufficient voltage to power the video camera, processing unit, wireless transceiver, display, horizontal and vertical light emitting diodes (LEDs).

[0064] In one embodiment, the system is edge-capable and can operate with or without an internet connection, which enables software and database updates.

[0065] In one embodiment, the present invention relates to a system and method for authenticating gemstones, and more particularly, to a portable, easy-to-use device that uses a video camera inspection system to authenticate loose stones and gemstones in jewelry through video-to-image matching.

[0066] The present invention generally addresses the needs mentioned in the previous section by providing methods and systems for identifying loose stones and gemstones in jewelry. Embodiments of the systems and methods will be further described in detail below.

[0067] One embodiment of the present invention provides a novel method for gemstone identification that focuses on analyzing the unique identifying marks and physical characteristics of each gemstone. This method is designed to meet the needs of various stakeholders in the gemstone industry with a highly automated approach. The identification process utilizes a series of artificial intelligence algorithms—namely, AI detection and recognition algorithms, AI edge detection algorithms, and AI matching algorithms—that work together to capture, process, and match gemstone features extracted from streaming digital video footage.

[0068] In another embodiment of the present invention, the method comprises a comprehensive process that includes detecting and identifying gemstones and their unique markings using an AI-based gemstone and unique marking detection and identification system, feature extraction via an AI-based edge detection system, and AI-based feature matching. The method includes a description of a training and data augmentation process for improving the algorithm's robustness to variations in images captured in a digital video stream. The method also provides a method for filtering image noise during the matching step using features extracted from a reference image. The final step in the authentication process involves calculating a similarity score, which is then compared to a predetermined threshold to determine the gemstone's authenticity. If there is more than one image in the reference atlas that meets the criteria, the process is repeated.

[0069] By employing AI-based techniques for edge detection and feature matching, the method aims to overcome challenges associated with image noise, variations in image quality, and other factors that can affect the accuracy of the identification process. Automating the identification process minimizes the need for human intervention in these critical steps, traditionally performed manually by gemologists.

[0070] In another embodiment of the present invention, the system is a device for gem identification, which allows users to control and capture streaming video through physical interaction. The system is also capable of processing, transmitting, and receiving data through the Internet and intranet, and communicating with cloud servers. Embodiments of this system include but are not limited to those described in the specification. In one embodiment, the system of the present invention can be deployed in multiple locations in different stores. Each local system includes a high-resolution digital video camera, a processing unit, a wireless Internet and intranet connection module, and a display. Each system is operated by a target user to operate a video camera module to capture streaming video of the target gem. When a gemstone and an inscribed certificate number are detected, the system automatically processes and transmits the streaming video in real time. The streaming video is transmitted to the cloud via a wireless transmission module for video-to-image matching, and the matching results are returned.

[0071] To further illustrate the cloud-compatibility of the present identification method, one embodiment of the present invention receives processed streaming video from a cloud server via the system's wireless transceiver module. The server module hosts the video-to-image matching process and stores a database of real gemstone images. Upon request from any local edge device, the video-to-image matching process is executed. Upon further request from any local system, the gemstone identification results are provided to the corresponding local system via the internet or intranet.

[0072] The present invention provides an adjustable workbench that can hold gemstones in place for easy video recording. The height and orientation of the workbench can be adjusted to allow viewing of gemstones from different perspectives during the identification process.

[0073] The present invention also provides a battery device and a wireless Internet and intranet connection module. These technical features eliminate the need for wired network connection, and coupled with the lightweight and compact design, enable the portability of the system.

[0074] In summary, the present invention provides a gem identification system and method that utilizes artificial intelligence algorithms. The system is highly automated and easy to use, requiring no extensive training. The specification also describes additional benefits and features achieved by the system of the present invention. These features are further described in the detailed description and accompanying figures.

[0075] The purpose of the following specific embodiments is only to illustrate the concept of the present invention so that those skilled in the art can more easily understand it. The present invention is not limited to the specific technical solutions described in the embodiments, but includes all technical solutions within the scope of protection of the claims. In this application, various references or publications are cited. The disclosures of these references or publications are incorporated into this application by reference in their entirety to more comprehensively describe the prior art in the field to which the present invention belongs. It should be noted that the transitional term "comprising" is synonymous with "including", "comprising" or "characterized by", is inclusive or open-ended, and does not exclude additional, unlisted factors or method steps.

[0076] The gem identification system of this invention is designed to meet the demand for portability, ease of operation, and edge capabilities. Specifically, the system can capture and process streaming video of gemstones, automatically detect physical features and markings, such as inscription information and certificate numbers, and perform and display the identification results using video-to-image matching technology. Ease of use also requires no tools and can be carried by a typical store assistant, automatically performing the identification process. Furthermore, edge capabilities utilize video-to-image matching technology to perform computational and storage requirements.

[0077] Figure 1 Provide a process description of the identification method; Figure 2 Describe the overall user operation process of the authentication system. Figure 3 Describe the functional modules within the system. The method of the present invention will first be described independently of the user operation process, and then described in conjunction with the system user operation process. In addition to the description of precautions, both the method and the user operation process will include references to the system functional modules.

[0078] The method of the present invention relates to the identification of gemstones in loose stones and jewelry items, wherein the gemstones include at least one unique identifying mark or characteristic or both. Figure 1As shown, once the identification process begins using the system of the present invention, the high-resolution digital video capture device 110 captures video of the gemstone being inspected. At a given time, one frame of the streaming video is sampled as an image, hereafter referred to as the "detection image," and a custom weighted transformation is performed on the image's color components. In process 502a, an artificial intelligence detection and recognition algorithm running on the local processor unit 280 is used to detect the gemstone and its corresponding unique identifying mark. During detection, the algorithm identifies the unique identifying mark or feature, or both, and returns the corresponding machine-readable identifier in the form of a string to 502b. The artificial intelligence detection and recognition algorithm, trained on images of the mark and gemstone, along with data augmentation, allows the algorithm to detect and identify the gemstone and its unique identifying mark or feature, or both, from digital video input images during inference, thereby improving its robustness to variations in image quality, mark shape, distortion, and lighting conditions. Once the gemstone and its corresponding unique identifying mark or feature, or both, are detected and identified, the corresponding image, including the identified machine-readable mark, is sent to a cloud server for further processing in the identification process. During the inference process, it's possible that nothing is detected. If the gemstone and its corresponding unique identifying mark or feature, or both, are not detected within the time limit, the system will time out 407, thus adding a layer of protection against misuse of system resources. The detected mark is used to generate a machine-readable text string that is used to retrieve the corresponding gemstone reference image set.

[0079] When the cloud server receives a test image with a gemstone and corresponding markings, the server will retrieve a reference image set based on the markings. The artificial intelligence edge detection algorithm 503 will then extract edge features from the test image and the reference image set. The edges identified in the gemstone image include the edges of the cutting facets and girdle, the edges of unique marks and physical features. From these edge features, traditional features such as cutting facet proportions, cutting facet angles, cutting facet dimensions, cutting facet intersections, girdle thickness, girdle finishing, inclusion features, inscribed information ratios, certificate numbers, unique identification numbers, and custom markings can all be inferred. In addition, the power of artificial intelligence edge detection lies in its ability to extract deeper features without the need for parameter operations, as well as its robustness to changes. This is based on training this artificial intelligence edge detection algorithm with a reference image set and its corresponding annotated edges. In order to improve the algorithm's robustness to changes in image attributes, data augmentation of the training images is introduced. Data augmentation includes but is not limited to changes in contrast, brightness, rotation, shift, and miscut. The introduction of data augmentation enables the artificial intelligence edge detection algorithm to adapt to various changes. The edges extracted from the test image are as follows: Figure 4 and Figure 5 The feature extraction method is applied to both the detection image and the reference image set to prepare for the subsequent matching process 504a.

[0080] Once extracted, the aligned edge features are passed to an artificial intelligence matching algorithm for the feature matching process 504a. The matching algorithm is trained based on the edge features extracted from the original and data augmented reference images. Therefore, it is able to learn the true edge features from the reference image set. Once the edge features extracted from the test image are received, the edge features of the test image are first affine transformed and aligned with the edge features in the reference image set to eliminate any matching errors caused by the difference in perspective between the test image and the reference image. During the alignment, the edge features from the test image are compared with the edge features of the corresponding reference image set and a similarity score is calculated for the matching results 504b. The similarity score represents the confidence of the algorithm as to whether the test image and the reference image set match. The similarity score ranges from 1 to 0, where 1 is a confident match and 0 is a confident mismatch, and the score can be easily adjusted to any value within this range. A common difficulty in the matching process is noise in the inspection image caused by dust particles, such as Figure 4 As shown, the dust particle outline is also identified as an edge feature. The method of the present invention allows the artificial intelligence matching algorithm to distinguish and ignore these noise edge features during the matching process. This method uses the edge features of the reference image as the edge features extracted from the detection image for hard filtering. Among the edge features of the reference image, only the edges that exist in the reference image are retained, thereby eliminating any noise edge features that should not be considered in the matching process. Figure 5 In the case of mismatches shown, these edge features that are not present in the reference image set are filtered out from the edge features extracted from the detection image, resulting in a low similarity score. This allows the algorithm to handle changes in capture angle, distortion, and image noise without affecting matching accuracy. After receiving features from the new detection image and the reference image set, the algorithm performs matching and outputs a similarity score for subsequent processing.

[0081] The similarity score is passed to a preset similarity threshold 505. This similarity threshold is set based on the domain knowledge and security requirements of the entity operating the present invention and can be a value between 0 and 1, or arbitrarily determined within the corresponding similarity score range. The similarity threshold is used to determine whether the gemstone being inspected matches the reference image in the database. Operating the system with an obvious example of extremely low tolerance for counterfeiting will set the threshold as close to 1 as possible. If the similarity score is greater than or equal to the predetermined similarity threshold, the gemstone being inspected will be considered to match the database record. On the other hand, considering the situation where the similarity score is less than the preset similarity threshold, the gemstone being inspected will be considered to have failed the inspection, and the result will be that the gemstone being inspected does not match the database record. The similarity threshold can be adjusted based on the domain knowledge and security requirements of the user of the present invention.

[0082] Figure 4 A process description outlining the methods of the present invention is provided. Figure 4 Displays successful identification results that match the gemstone sample under test with the database reference image record. First, the artificial intelligence edge detection algorithm extracts edge features from the "reference image with unique markings" to obtain the "edge features extracted from the reference image". This process is then repeated for the "test image with dust particles" to produce the "edge features extracted from the test image". Note Figure 4 The edge below the letter "V" in the image has noise caused by dust particles. The edge features extracted from the test image are filtered using the edge features extracted from the reference image to eliminate this noise, resulting in the filtered test image edge features. A similarity score is calculated by matching the uniquely marked reference image with the filtered test image edge features using an artificial intelligence matching algorithm. The algorithm correctly identifies and matches the edge features extracted from the images, resulting in a high similarity score close to 1. A gemstone tested and its corresponding record in the database are considered a match if the calculated similarity exceeds the user-defined threshold. Figure 5 A process description is provided that further outlines the methods of the present invention. Figure 5 Displays the identification result that the tested gemstone sample does not match the reference image record in the database. Figure 4 The process is similar to that of , and the similarity score is calculated based on the edge features of the current "filtered detection image" and the "reference image with unique markings" through the artificial intelligence matching algorithm. Figure 5 In the example above, the algorithm correctly identifies the mismatched features: the girdle of the gemstone in the test image is thinner than that in the reference image. Therefore, the calculated similarity score is close to 0. If the similarity score fails to pass the user-defined similarity threshold, the tested gemstone will be considered a mismatch with the corresponding record in the database.

[0083] In a simplified extension of the above invention, multiple reference images are added to each gemstone in the reference set. The same process is repeated for each individual reference image. Each reference image in the reference set is scored based on a similarity score with the captured test image. If all similarity scores exceed a threshold, the test is successful and the gemstone is deemed authentic. If any of the similarity scores fail to exceed the threshold, the test fails.

[0084] When verifying the performance of the method of the present invention, the effectiveness of the steps mentioned in this method is demonstrated by using the verification results of real gemstone data. Figure 7The results of a diamond match test against 95,638 real diamonds in the database. This method was able to correctly identify the tested diamond from the database with a false positive rate of 0%. Figure 6 is the test result of the diamond's match to the database record. The test image is paired with the corresponding reference image in the database record. Each pair serves as a sample in the experiment. Therefore, a predicted positive sample indicates that the method correctly identified each pair as a match, while a predicted negative sample indicates that the method incorrectly identified the sample as a mismatch. This method achieved an accuracy of 99.6% based on a dataset of 500 experimental pairs. This experiment validated the method's ability to correctly identify each pair of test and reference images.

[0085] The following demonstrates the necessity of the steps in the present method. Experimental results using real diamond samples and data further validate the present method. After removing some key steps from the present method and performing the matching, the accuracy of the test results demonstrates that these key steps are essential.

[0086] In validating the artificial intelligence detection and recognition algorithm in this method, the algorithm includes a method for performing image correction using key landmarks in the image. Figure 6 The experimental data in

[15] confirm the importance of using key markers in image correction. When image correction is performed without using key markers, the matching accuracy is reduced to 0% according to the experimental data.

[0087] To further demonstrate the method of the present invention, diamonds matched in the database and reference images were used to verify the method of using reference images for hard filtering in the artificial intelligence matching algorithm. Figure 6 This proves the effectiveness of hard filtering with reference images. Without hard filtering with reference images during the matching process, the matching accuracy drops to 0% again.

[0088] The method is robust and can operate even in storefronts with extreme lighting conditions. Experiments have also demonstrated that it can operate at very low levels of contrast and brightness. To test the method's capabilities under different lighting conditions, the test images in each pair were subjected to varying percentages of contrast and brightness changes. Brightness and contrast were reduced by 60%, 40%, 20%, and increased by 20%. Figure 8 Shown below are examples of inspection images resulting from this transformation. The brightness and contrast values chosen above simulate the extreme conditions that operators in a store setting might encounter, often in dimmer lighting conditions than in a lab environment. While brighter lighting conditions are possible in stores, they are less likely and easily correctable, hence the ranges selected. Figure 8 The results show that the accuracy of the method of the present invention is higher than 97% under different lighting conditions.

[0089] The present invention also relates to a system capable of implementing the identification method of the present invention. Figure 3 The modules within the hardware housing of the present system are shown. This gem identification system includes a high-resolution digital video camera 110. The horizontal position of the video camera is adjustable, allowing the streaming video to be better focused at the correct distance for gemstone inspection. The streaming video can be digitally zoomed in and out to facilitate inspection. The hardware housing 190 consists of a processor unit 280 and a wireless transceiver module 253, powered by a rechargeable battery unit 254 also housed within the housing. To provide optimal lighting conditions for capturing gemstone video, the system includes horizontal and vertical lighting devices. The horizontal lighting device 100 provides backlighting, while the vertical lighting device 101 is supported by a flexible cable assembly, allowing the present invention to provide illumination at different heights and angles through vertical and rotational adjustment. An adjustable work surface is used to place and video record loose stones and gemstones in jewelry. The item to be inspected is placed on the flexible support of the adjustable work surface, and the gemstone can be identified from different perspectives by changing its height and orientation.

[0090] The following parts refer to Figure 2 and Figure 3 To operate the system, the user will turn on the system via the power switch. The system will start in gemstone viewing mode 401. The user can then choose to enter identification mode 403 via the push button switch.

[0091] As described above, when switching to authentication mode 403, user verification and a software version check are first performed. Considering the situation of a new user, the user needs to perform the following steps to activate the system. The new user will be directed to device activation 400. In device activation 400, the user is required to scan a QR code displayed in the device activation user interface, where the user's identity will be verified through an external authentication system. Once verified, the device will be activated by adding a unique identification code to the database, and the user can proceed to the next steps. If the user has been verified and the system is already activated, device activation 400 will be skipped and the system will proceed directly to the next step, version update 402.

[0092] Version update 402 is an automated process. Wireless transceiver module 253 first transmits the system's software version to the cloud. The cloud server then executes a program to determine whether the software version is up to date. If the software is not up to date, the updated version is automatically installed on the system via wireless transceiver 253. If the latest software is already available or there is no internet connection, the new software will not be received or installed.

[0093] Beginning with version update 402, the processor module 300 and memory module 301 of the system processor unit 280 within the hardware housing 190 work together to automatically process the streaming video captured by the high-resolution digital video device 110. The processing unit 280 will then execute the gemstone physical feature and certificate number inscription detection loop 404 in real time. The user can adjust the viewing angle of the gemstone and the laser inscription by manipulating the adjustable work table and focus adjustment knob. The user can also adjust the magnification of the streaming video on the display 150 to suit their needs.

[0094] Once the gemstone's physical features and certificate number inscription are detected, the processed video can be transmitted to the cloud or to the local processor unit 280 for video-to-image matching. If the processed video is transmitted to the cloud, the transmitted data is encrypted during transmission and decrypted on the cloud server. As described in process 404, an automatic time limit test is performed in the loop process. If the gemstone and certificate number inscription are not detected, the system will proceed to "Timeout" 407 and the display 150 will display the timeout interface. If desired, the user can restart the identification process by pressing the "Start" button shown in the user interface on the display 150.

[0095] If the aforementioned artificial intelligence detection and recognition algorithm is used to detect and identify gemstones and their unique markings, the processed video will be transmitted to the cloud via the wireless transceiver module 253. The cloud server will perform the automatic video-to-image matching process 106 described in the present invention based on a cloud database storing gemstone images.

[0096] The cloud database on the cloud server includes encrypted sets of gemstone images for executing the video-to-image program 408. Each set of gemstone images belongs to a specific authenticated gemstone. Each set of gemstone images may include one or more pictures. Each set of images includes at least one picture that carries a unique identification mark or feature, or both, for identifying the gemstone.

[0097] If the gemstone is determined to be authentic by matching the gemstone image records in the database 406, the display module 150 will display an "Identification Passed" page confirming the gemstone's authenticity. If the gemstone is found to be counterfeit or cannot be matched with any gemstone recorded in the database 405, an "Identification Failed" page will be displayed, indicating that the gemstone does not match the database record, thus ending the automatic identification process. The automatic identification process is crucial to the functionality of the present system because it improves the system's ease of use and reduces the training required for non-gemologists, further reducing the time and human error involved in the gemstone identification process. In one embodiment, the system can be operated by sales staff in a jewelry store.

[0098] In addition, the gem identification system includes a battery module 254 for powering the system, which enables the system to be portable and, together with the wireless transceiver module 253, eliminates the need for wired connections. The compact and portable design of the gem identification system makes it easy to transport and deploy in stores.

[0099] As shown above, the gem identification capabilities of this system can be performed regardless of whether it is connected to the Internet. Combined with the system's automated identification process and portability, this system is suitable for, but not limited to, deployment in a store or laboratory environment.

Claims

1. A computer-implemented method for identifying a specific gemstone having a known reference point, characterized in that: The following steps are involved: a. detecting the known reference point from a source image of the particular gemstone; b. retrieving one or more reference images of a possible gemstone from a gemstone database based on the known reference points; c. extracting edge features from the source image and the one or more reference images; d. calculating a similarity value by pairing the edge features of the source image and each of the one or more reference images, wherein the pairing includes hard filtering the source image using the edge features from the one or more reference images; and e. Identifying the specific gemstone from the one or more reference images based on the similarity value.

2. The method according to claim 1, wherein: The known reference point is an inscribed mark or a natural mark.

3. The method according to claim 1, wherein: The source image is a still image from a camera or an image obtained from streaming video processing.

4. The method according to claim 1, wherein: One or more of steps (a), (b), (c) or (d) are performed using an artificial intelligence algorithm.

5. The method according to claim 4, characterized in that: The artificial intelligence algorithm is trained with multiple labeled images of landmarks and gemstones along with data augmentation.

6. The method according to claim 1, wherein: One or more of steps (a), (b), (c) or (d) include filtering image noise.

7. The method according to claim 1, wherein: The known reference point is located on, near or outside the girdle of the particular gemstone.

8. The method according to claim 1, wherein: The step (b) includes generating a string of machine-readable text based on the record of the known reference points, and retrieving the one or more reference images recorded together with the string of machine-readable text.

9. The method according to claim 1, wherein: The step (d) comprises aligning edge features of the source image and the one or more reference images.

10. The method according to claim 1, wherein: The step (e) includes retaining only images whose similarity values exceed a threshold.

11. A non-transitory computer-readable storage medium having stored thereon instructions, characterized in that: When executed by one or more processors, causes the one or more processors to perform operations comprising the computer-implemented method of claim 1.

Images