SYSTEM AND METHOD FOR DETECTING AND CLASSIFYING PATTERNS IN IMAGES WITH A VISION SYSTEM - Patent application

Abstract

To provide a system and method for detecting patterns in images that incorporate neural net classifiers.SOLUTION: In a vision system 100, a pattern detection tool is coupled with a classifier 150 that can be run before or after the tool to have labeled pattern results with sub-pixel accuracy.EFFECT: In the case of a pattern detection tool that can detect multiple templates, its performance is improved when a neural net classifier informs the pattern detection tool to work only on a subset of an originally trained templates. In the case of a pattern detection tool that initially detects a pattern, a neural net classifier can then determine whether the tool has detected a correct pattern, can also reconstruct / clean-up an imaged shape, and eliminate pixels less relevant to the shape of interest, therefore reducing search time, as well significantly increasing the possibility of lock on correct shapes.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to machine vision systems and methods, and more particularly to pattern searching and recognition tools. [Background technology]

[0002] Machine vision systems, also referred to herein as "vision systems", are used to perform a variety of tasks in manufacturing environments. Generally, a vision system consists of one or more cameras with image sensors (or "imagers") that capture grayscale or color images of a scene containing the object being manufactured. The images of the object can be analyzed to provide data / information to a user and the associated manufacturing process. Data generated by the images is typically analyzed and processed by the vision system in one or more vision system processors, which may be purpose-built or part of one or more software applications instantiated within a general-purpose computer (e.g., PC, laptop, tablet, or smartphone).

[0003] Common vision system tasks include alignment and inspection. In alignment tasks, vision system tools, such as the well-known PatMax® system available from Cognex Corporation of Natick, Massachusetts, compare features in an image of a scene to trained patterns (using real or synthetic models) to determine the presence and pose of the pattern in the imaged scene. This information can be used in subsequent inspection (or other) operations to search for defects or to perform other operations such as rejecting the part.

[0004] It is desirable to improve the performance of conventional pattern detection tools that can include a predefined list of searchable patterns (e.g., images of circles, squares, screws, etc.) Such tools often fail to adequately distinguish between certain trained patterns that have subtle differences (e.g., a circle and a circle with a small notch). Summary of the Invention

[0005] The present invention overcomes the shortcomings of the prior art by providing a system and method for detecting patterns in an image that incorporates a neural network (also called "neural net") classifier (sometimes called "analyzer"). The pattern detection tool is coupled with a classifier that can run before or after the tool to have the pattern detection results labeled with sub-pixel accuracy. For pattern detection tools that can detect multiple templates, the performance is improved if the neural net classifier informs the pattern detection tool to work only with a subset of the templates it was initially trained on. Alternatively, the performance of the pattern detection tool can be improved if a neural network is used to reconstruct / clean the image before running the pattern detection tool. Additionally, a neural network can be used to calculate a weighting value for each pixel in the image based on the likelihood that the pixel belongs to the pattern and matches. Similarly, for tools that initially detect patterns, the neural net classifier can determine whether the tool has found the correct pattern or not.

[0006] In an exemplary embodiment, a system and method for detecting a pattern in an image includes a pattern detection tool trained based on one or more templates associated with one or more training images containing a pattern of interest. A neural net classifier is trained on the one or more training images, and a template matching process operates at run-time such that (a) the trained neural net classifier provides one or more templates based on the run-time images to the pattern detection tool, and the trained pattern detection tool performs pattern matching based on the one or more template images combined with the run-time images, or (b) the trained pattern detection tool provides patterns detected from the run-time images to the trained neural net classifier, and the trained neural net classifier performs pattern matching based on the detected patterns and the run-time images. The pattern detection tool is adapted to be trained using multiple templates or a single template. The neural net includes a convolutional neural network (CNN).

[0007] In another embodiment, a system and method for detecting a pattern in an image is provided, comprising a neural network trained to identify one or more candidate shapes in an image and configured to identify a probability of the presence of one or more shapes in the image during run-time operation. The neural network thereby generates (a) a weighted mask having features of the one or more candidate shapes that exceed a probability threshold, and / or (b) a reconstructed image in which features of models of the one or more candidate shapes are replaced with an image in which the neural network analyzer identifies the presence of the one or more candidate shape features that exceed the probability threshold. Illustratively, the pattern detection tool is trained to detect one or more candidate shapes in (a) the weighted mask and / or (b) the reconstructed image using one or more models associated with the one or more candidate shapes. The neural network defines a weighted mask in which each pixel has a score related to the identification of the one or more shapes. The reconstructed image can be defined as a binary image. Illustratively, the neural network analyzer provides data regarding the presence of one or more types of candidate shapes to the pattern detection tool, which limits the process to those related to identifying the types. Generally, the neural network may include a convolutional neural network (CNN).

[0008] The present invention will now be described with reference to the accompanying drawings. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram of a vision system configured to analyze imaged objects containing different shapes and / or patterns using pattern detection tools combined with a neural network classifier.

[0010] [Diagram 2] FIG. 1 is a block diagram illustrating a training procedure for a smart pattern detection tool that includes a neural network classifier and a pattern detection tool trained on the same set of one or more image templates.

[0011] [Diagram 3] FIG. 3 is a block diagram of run-time operations on an input image using the trained smart pattern detection tool of FIG. 2, where a neural net classifier runs before the pattern detection tool.

[0012] [Figure 4] FIG. 1 is a block diagram illustrating a training procedure for a smart pattern detection tool that includes a neural network classifier and a pattern detection tool trained on one or more images for a single template.

[0013] [Diagram 5] FIG. 5 is a block diagram of run-time operations on an input image using the trained smart pattern detection tool of FIG. 4, with a neural net classifier running after the pattern detection tool.

[0014] [Figure 6] 1 is a flowchart outlining a procedure for first applying a neural network to search / decide candidates in an image and then applying a pattern detection tool to refine the search / decide.

[0015] [Figure 7] 7 is a flow chart illustrating the training and run-time operations of the neural network in the procedure of FIG. 6.

[0016] [Figure 8] 8 is a flow chart illustrating the run-time operation of a pattern detection tool that uses image results provided by the neural network procedure of FIG. 7.

[0017] [Figure 9] FIG. 6 is a block diagram illustrating the creation of a weighting mask associated with the pattern detection tool of FIG. 5.

[0018] [Figure 10]10 illustrates an exemplary shape of interest and processing of acquired image data including a shape of interest and a model of the shape of interest to generate a weighted mask according to FIG. 9.

[0019] [Figure 11] FIG. 6 is a block diagram illustrating the use of a neural network according to FIG. 5 to reconstruct or remove features of interest in acquired image data.

[0020] [Figure 12] FIG. 12 illustrates the processing of exemplary acquired data including a feature of interest to generate a reconstructed and / or removed version of the feature (eg, a binary image) according to FIG. 11 . DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] I. System Overview

[0022] FIG. 1 illustrates a generalized vision system configuration 100 for use with exemplary systems and methods. The vision system can be implemented in any permissible environment, including part / surface inspection, robotic control, part alignment, and the like. The system includes at least one vision system camera assembly 110 having an optical system O and an image sensor (also called a "sensor" or "imager") S, and may also include on-board lighting or separate lighting (not shown). The camera assembly images a scene 120 that may include one or more stationary or moving objects 130. In this example, the objects include contour shapes, as well as a variety of geometrically regular or irregular interior shapes 132, 134, 136, and 138. More generally, the pattern may be any 2D image of any geometric shape or object.

[0023] The camera assembly 110 and associated sensors S are interconnected with a vision system processor 140, which may be located wholly or partially within the camera assembly 110, or may be located within a separate processing device, such as a server, PC, laptop, tablet or smartphone (computer 160). The computing device may include a suitable user interface, such as a display / touch screen 162, a keyboard 164 and a mouse 166.

[0024] Illustratively, the vision system process (processor) 140 operates various vision system tools and associated software / firmware to process and analyze images of the object 130 acquired and / or stored during execution. The process (processor) 140 functions according to certain parameters and can be trained to recognize certain shapes detected on the object using a training procedure. The process (processor) 140 includes various vision system components, including a pattern detection tool 142 as found in the PatMax® software package and its variations, such as PatMax® Multimodel, mentioned above. The pattern detection tool can use trained patterns and patterns of standard shapes (squares, circles, etc.), which are included in the training templates. As described below, the vision system process (processor) also includes or works in conjunction with a neural network process (processor) 150. The neural network process (processor) (also called a "neural net") operates on various patterns in the form of a classifier to enhance the pattern detection speed and performance of the system 100.

[0025] The results of the pattern detection can be transmitted via computer interface 162 to a user and / or other downstream utilization devices or processes (processors) 180; such devices or processes (processors) may include assembly robot controllers, line inspection, part inspection / rejection, quality control, etc.

[0026] It is recognized that conventional approaches to pattern matching involve training a conventional pattern matching tool, such as PatMax® or PatMax® MultiModel, with model images having shapes or features of interest. At run-time, the pattern matching tool runs through one or more (possibly all) of the trained templates in an effort to identify an exact match with the trained pattern in the acquired images of the object being inspected.

[0027] On the other hand, the present embodiment provides a smart pattern detection tool that utilizes neural net processes to augment conventional pattern detection tools, thereby providing the pattern detection tool with the ability to automatically label detected patterns in the tool results or reliably detect patterns using an associated neural net classifier. This approach allows the smart pattern detection tool to be trained in operation on a database of images containing templates. Once trained, at run time the smart pattern detection tool combines the best features of conventional pattern detectors and the neural net classifier to provide correctly labeled pattern detection results with high accuracy of pose (location, scale, rotation, etc.).

[0028] II. Pattern detection using neural networks to refine searches

[0029] 2, which shows a block diagram illustrating the training-time procedure 200 prior to run-time. The tool 210 includes a conventional pattern detection tool 220 and an associated neural net classifier that assists the pattern detection tool in detecting the correct pattern among a set / multiple trained patterns (templates) 240.

[0030] More specifically, during training, a conventional pattern detection tool 220 (e.g., PatMax® Multi-Model) is trained on one or more template images. At the same time, a neural net classifier (e.g., a Convolutional Neural Network (CNN)) 230 is trained on multiple example images of the patterns represented by each template. The neural net classifier 230 is trained to process the input images and report a subset of template labels detected in the input images.

[0031] FIG. 3 shows a run-time procedure 300 (using pre-classification), where a trained neural network classifier 310 is first run on a run-time input image 320 (acquired by the camera 110 and / or saved from a previous acquisition) to determine the probability of each template. The smart pattern detection tool filters the best results and then informs the trained pattern detection tool 350 to work with a subset of these matched M templates instead of the full set of N templates (240 in FIG. 2). Thus, the subset of the best matching M templates 340 is provided as search parameters to the trained pattern detection tool 350. The input image 320 is provided to the pattern detection tool 350, and the search parameters are used to generate matched template results 360 as the output of the tool 350. These results can be displayed to the user or used in downstream exploitation operations (e.g., part alignment, inspection, etc.). Advantageously, this process 300 reduces the amount of computation that would otherwise be required to generate matched templates.

[0032] Various proprietary and commercially available (e.g., open source) neural network architectures and associated classifiers can be used in accordance with embodiments herein, such as TensorFlow, Microsoft CNTK, etc.

[0033] An example application in which the above-described training-time procedure 200 and run-time procedure 300 can be used is to find the correct fiducial when the shape of the fiducial can vary with different parts (e.g., cross, diamond, etc.). Illustratively, a conventional pattern detection tool is trained with template images representing each possible fiducial pattern. In addition, a neural net classifier (e.g., tensorflow) is trained with multiple images showing the variations in appearance of each fiducial pattern, along with the labels associated with each fiducial pattern. At run-time, the first trained neural net classifier is executed to return a set of labels detected in the run-time images. Using this information, the system informs the pattern detection tool (e.g., PatMax® Multi-Model) to run only on the set of templates represented by the labels generated by the neural net classifier, thereby speeding up the alignment and producing more reliable results.

[0034] In the configuration shown in Figure 4, a conventional pattern detection tool detects a particular pattern and a neural net classifier determines whether it is a correct match (i.e., post-classification). During training, a conventional pattern detection tool 410 is trained with a single image template 440. A neural net classifier (e.g., tensorflow) 430 associated with the tool 410 is then trained with multiple images for the desired template. The input for the classifier 430 is the same image that was input to the conventional pattern detection tool 440.

[0035] 5 shows a run-time procedure 500 (using post-classification) where a pattern detection tool 550 first detects a pattern in an input image 520. The input image 520 is provided to a trained neural network classifier 510 along with an optional bounding box calculated from the output of a conventional pattern detection tool (e.g., PatMax®). The classifier 510 then determines whether the conventional pattern detection detected the correct / matching pattern 560. The output 510 of the classifier is an overall confidence that the trained template was detected.

[0036] As an example, an operational use case involves very confusing shapes with slight differences, such as a circle and a circle with a notch. Assume that a conventional pattern detection tool (e.g., PatMax®) 350 is trained with a template image representing a circle with a notch. A neural net classifier 510 is then trained with images containing the desired shape (a circle with a notch) along with other confusing shapes (circles without notches). At run-time, an input image is fed to the trained neural net classifier 510 along with an optional bounding box calculated from the output of the conventional pattern detection tool, and the classifier then determines if the conventional pattern detection detected the correct pattern (a circle with a notch). This procedure improves the robustness of the pattern detection in this exemplary case.

[0037] It should be noted that the conventional pattern detector and its ability to be trained with one or more templates is highly variable in alternative embodiments. The pre-classification and post-classification procedures described above can each be modified to include different types of pattern detectors and associated templates in alternative embodiments.

[0038] III. Pattern Discovery: Refine your search using trained pattern tools

[0039] Referring to FIG. 6, an overall (generalized) procedure 600 for smart pattern detection according to another exemplary embodiment that can be implemented by the configuration 100 of FIG. 1 is shown. It is considered that some patterns may prove more difficult to identify using a conventional trained pattern detection tool operating on the acquired image. In some cases, the inherent properties of neural network / deep learning architectures may provide advantages in initially identifying pattern candidates in the image. Thus, in procedure 600, a neural network is trained to identify various pattern types and is applied to the acquired image in step 610, thereby generating a list of candidates with scores associated with a given type of shape. Then, based on the scores, procedure 600 applies a conventional pattern detection tool (e.g., PatMax® Multi-Model) to the shape candidates with scores above a certain threshold (step 620). The pattern detection tool may search for the specific shape identified by the neural network or for various types of shapes in each candidate.

[0040] Advantageously, neural networks can efficiently identify possible candidates, while computationally demanding tasks such as model fitting at the sub-pixel level can be handled in a robust manner by pattern detection tools.

[0041] Training the neural network to recognize a particular shape is described in step 710 of procedure 700 of FIG. 7. Once trained, at run time (step 720), the neural network uses the trained configuration to assign a score (probability) to each pixel of an acquired image based on whether the image appears to be part of the trained shape. The result (step 730) is a probability image in which each pixel in the image has been assigned a score. The probability image from step 730 can be saved and then provided to a pattern detection tool (e.g., as a mask). In the pattern detection tool, pixels that do not appear to have a candidate shape are masked out of the image results on which the pattern detection tool operates (step 740). The results of the neural network may include the type of the candidate shape in the probability image. The shape type information allows the pattern detection tool to narrow its search (at selected locations) in the probability image exclusively to the shape type provided by the results (step 750). The pattern detection tool can therefore operate more quickly and efficiently, since it avoids running tools that are not related to the candidate shape.

[0042] The above procedure 600 is advantageous in various applications. For example, using a neural network to initially screen the image is useful when local distortions are large, because the neural network essentially reconstructs the image based on probabilities in a way that is easier to analyze by pattern detection. As an example, the incoming image may be highly textured and lacking defined contrast lines. After processing by the neural network, the resulting probability image is a binary representation with high contrast defined boundaries representing rectangles, triangles, circles, etc. In a specific example, the neural network can effectively resolve shapes at the ends of a rope or cable that may be frayed (creating highly textured areas). The neural network provides the pattern detection tool with light rectangles on a dark background, or vice versa.

[0043] As described in step 800 of Figure 3, at run time, a trained pattern detection tool (trained with a model-based template for shapes of interest) receives a probability image (mask) from the neural network, and (optionally) information regarding the type of candidate shape identified in the image (step 810). The pattern detection tool focuses on a selected region on the image and operates using tools and processes relevant to the identified image type (step 820). The pattern detection tool then generates a result in which the detected shapes are located in the image, and appropriate coordinate (and other) data for the shapes is output for subsequent operations in step 830.

[0044] Further referring to FIG. 9, a block diagram of an exemplary procedure 900 for creating and using a weighted mask in relation to a shape of interest in a captured image is shown. As shown therein, an image 910 is input to a neural network 920. Using a properly trained technique, the neural network 920 outputs a weighted mask of the shape of interest 930. As described above, each pixel is scored based on the likelihood that it is part of the shape of interest. The weighted mask 930 is then input along with the raw image data 910 to a pattern detection (template matching) tool (e.g., Cognex SearchMax®, PatMax®, etc.). This causes the tool 940 to output additional matching score information 960, which may be based on the location of the shape of interest in the image 950 and the data contained in the weighted mask 930.

[0045] The procedure 900 is illustrated in the diagram 1000 of FIG. 10. An exemplary shape of interest 1010 is shown as a U-shaped structure with a continuous boundary. However, the associated acquired image 1012 provides a broken boundary 1014 and an intervening shape 1016. Furthermore, the image shape of interest 1012 is rotated at an angle relative to the expected orientation of the shape 1010 within the scene. There may also be other distortion-based differences between the acquired image and the expected shape. As described herein, the shape of interest data 1010 and the image data 1012 are input to a neural network 1020. The resulting image output weighted mask 1030 is represented as a series of shape segments 1040 that approximate the underlying shape of interest and omit the intervening shape data 1016. As shown here, the segments 1040 encompass a range of surrounding pixels where the probability / likelihood is higher. This region approximates the general contour of the edge of the shape of interest. This representation 1040 is more easily matched by conventional model-trained pattern detection (template matching) tools.

[0046] In another exemplary embodiment, a neural network can be used to reconstruct and / or remove shapes of interest in an image. As shown in step 1100 of FIG. 11, a neural network 1120 receives acquired image data 1110 and uses training to output a reconstruction of the shape of interest 1130, in which each pixel is scored by the likelihood that it belongs to the shape of interest (i.e., the subject of the neural network training). This reconstruction is then input to a model-based pattern detection (template matching) tool 1140 that contains a template of the shape of interest. The tool outputs a coarse location of the shape of interest 1150. This coarse location can be used by downstream processes, if applicable, and / or can be input to a pattern detection tool 1160 (either the same tool as block 1140 or a different tool), optionally again model-trained. The raw image data 1110 is also provided to the pattern detection tool 1160. The output of the tool 1160 from the inputs 1110 and 1150 is a fine location 1170 of the shape of interest in the image 1110.

[0047] As an example of procedure 1100, diagram 1200 in FIG. 12 shows two input shapes 1210 and 1212 in an image. Each shape undergoes neural network reconstruction 1220 and 1222 as described above, thereby producing reconstructed shapes 1230 and 1232, respectively, for use in the acquired image data. As a result, the reconstruction can replace an existing distorted or unclear shape. Thus, a neural network can be used to effectively provide removal and / or reconstruction of incomplete or distorted shapes in the image data, such data can be more effectively used by downstream operations, including pattern detection using the pattern detection tools described above or another suitable tool. As shown here, the shapes can be represented as binary images with well-defined boundaries that match the boundaries of the expected / model shape.

[0048] IV. Conclusion

[0049] It will be apparent that the above-described system and method provide a more reliable and faster technique for detecting and matching patterns trained using a combination of traditional pattern matching applications and neural net classifiers. This approach allows the number of templates to be reduced or the detected patterns to be filtered, thereby enhancing the system and method's determination of the correct match. Furthermore, the above-described system and method effectively allows the neural network to be used as a reconstruction / removal tool for the imaged shape and / or to eliminate pixels that are less related to the shape of interest, thereby significantly reducing the search time as well as significantly increasing the chances of locking on to the correct shape. This technique is particularly effective when the shapes in the image are distorted or feature-less.

[0050] The above is a detailed description of exemplary embodiments of the present invention. Various modifications and additions may be made without departing from the spirit and scope of the present invention. Features of each of the various embodiments described above may be combined with features of other described embodiments, where appropriate, to provide multiple feature combinations in related new embodiments. Moreover, while a number of separate embodiments of the apparatus and method of the present invention have been described above, what has been described herein is merely illustrative of the application of the principles of the present invention. For example, it should also be noted that the terms "process" and / or "processor" as used herein should be broadly interpreted to include a variety of electronic hardware and / or software based functions and components (or functional "modules" or "elements") that may be referred to as such. Furthermore, a process or processor as depicted may be combined with other processes and / or processors, or divided into various sub-processes or sub-processors. Such sub-processes and / or sub-processors may be combined in various ways in accordance with the embodiments described herein. Similarly, it is expressly contemplated that any functions, processes and / or processors herein may be implemented using electronic hardware, software, or a combination of hardware and software, comprising a non-transitory computer-readable medium of program instructions. In addition, the various directional and / or orientation terms used herein, such as "vertical," "horizontal," "up," "down," "bottom," "top," "side," "front," "rear," "left," "right," and the like, are used as relative descriptions only and do not represent absolute orientations with respect to a fixed coordinate system, such as the direction of gravity. In addition, when the words "substantially" or "approximately" are used in connection with a particular measurement, value, or characteristic, it refers to an amount that is within normal operating ranges to achieve a desired result, but includes some variation due to inherent inaccuracies and errors within the margin of error of the system (e.g., 1-5 percent). Therefore, this description should be taken only as an example, and is not otherwise meant to limit the scope of the invention.

[0051] The claims are set forth below.

Claims

1. 1. A system for detecting a pattern in an image, comprising: a neural network trained to identify one or more candidate shapes within an image and configured to identify a probability that one or more shapes of interest are present within the image during run-time operation, thereby (a) the neural network generates a weighted mask having pixels of the image that meet or exceed the likelihood of corresponding to the one or more shape features of interest; or (b) the neural network generates a reconstructed image, the reconstructed image including pixels that are determined by the neural network to meet or exceed a likelihood of belonging to the one or more shapes of interest. The above system.

2. 2. The system of claim 1, further comprising a pattern detection tool trained using one or more models corresponding to one or more of the shapes of interest to detect one or more of the shapes of interest in (a) the weighted mask or (b) the reconstructed image.

3. The system of claim 2 , wherein the neural network defines the weighting mask in which each pixel has a score related to an identification of one or more of the shapes of interest.

4. The system of claim 1 , wherein the reconstructed image is defined as a binary image.

5. 3. The system of claim 2, wherein the neural network provides data regarding the presence of one or more of the shape types of interest to the pattern detection tool, and the pattern detection tool limits the pattern detection process to those relevant to identifying the shape types of interest.

6. The system of claim 1 , wherein the neural network comprises a convolutional neural network (CNN).

7. 1. A system for detecting a pattern in an image, comprising: a neural network trained to identify one or more candidate shapes within an image and configured to identify a probability that one or more shapes of interest are present within the image during run-time operation, thereby (a) the neural network generates a weighted mask having pixels of the image that meet or exceed a likelihood of corresponding to the one or more shape features of interest; and (b) the neural network generates a reconstructed image, the reconstructed image including pixels that are determined by the neural network to meet or exceed a likelihood of belonging to the one or more shapes of interest. The above system.

8. 8. The system of claim 7, further comprising a pattern detection tool trained using one or more models corresponding to one or more of the shapes of interest to detect one or more of the shapes of interest in (a) the weighted mask and (b) the reconstructed image.

9. 1. A method for detecting a pattern in an image, comprising: using a neural network to find one or more candidate shapes within the image and identify a probability that one or more shapes of interest are present within the image during run-time operation; and (a) generating a weighted mask having pixels of the image that meet or exceed a likelihood of corresponding to the one or more shape features of interest; or (b) generating a reconstructed image, the reconstructed image including pixels that are determined by the neural network to meet or exceed a likelihood of belonging to the one or more shapes of interest; The above method comprising the steps of:

10. 10. The method of claim 9, further comprising detecting one or more of the shapes of interest in (a) the weighted mask or (b) the reconstructed image using a pattern detection tool trained with one or more models corresponding to the one or more shapes of interest.

11. The method of claim 10 , further comprising using the neural network to define the weighting mask such that each pixel has a score related to an identification of one or more of the shapes of interest.

12. The method of claim 9 , further comprising defining the reconstructed image as a binary image.

13. 13. The method of claim 12, further comprising providing data from the neural network to a pattern detection tool regarding the presence of one or more of the shape types of interest, and limiting a pattern detection process to that relevant to identifying the shape types of interest.

14. 10. The method of claim 9, wherein the neural network comprises a convolutional neural network (CNN).

15. 1. A system for detecting a pattern in an image, comprising: a neural network trained to identify one or more candidate shapes within an image and configured to identify a probability of the presence of one or more shapes of interest within the image during run-time operation, whereby the neural network generates a weighted mask having pixels of the image that meet or exceed a probability corresponding to features of the one or more shapes of interest; The above system.

16. 16. The system of claim 15, further comprising a pattern detection tool trained with one or more models corresponding to one or more of the shapes of interest to detect the one or more shapes of interest in the weighting mask.

17. 1. A system for detecting a pattern in an image, comprising: a neural network trained to identify one or more candidate shapes within an image and configured to identify, during run-time operation, a probability that one or more shapes of interest are present within the image, whereby the neural network generates a reconstructed image, including pixels within the reconstructed image that are determined by the neural network to meet or exceed a probability of belonging to the one or more shapes of interest; The above system.

18. The system of claim 17 , wherein the reconstructed image is defined as a binary image.

Images