A white eye choroid feature recognition method, device, computer storage medium and electronic equipment

Abstract

The application discloses a white eye choroid feature recognition method and device based on a human eye white eye visceral partition method, a computer storage medium and electronic equipment. The method comprises the following steps: acquiring a human eye image to be recognized; inputting the human eye image to be recognized into a first net structure model, calculating the choroid shape in the human eye image 、 Color and feature point coordinate information of the direction; the feature point coordinate information is classified and sorted to generate a feature point data unit, and the classification and sorting are based on the corresponding relationship between the feature point and the choroid; a second net structure model is used to calculate the human eye choroid features corresponding to the feature point data unit, and the human eye choroid features comprise color, thickness and spot blood vessel image information; and the feature point data unit and the human eye choroid features are used to generate a first image comprising at least human eye choroid feature information. The human eye white eye choroid feature recognition method provided by the application has the advantages of high recognition accuracy, small adaptive data volume, fast iteration in actual use, better universality and convenience for rapid production.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of human eye image recognition, and in particular to a white sclera choroid feature recognition method and device, a computer storage medium and an electronic device. BACKGROUND

[0002] In recent years, in the field of eye auxiliary information processing, how to automatically and accurately extract effective features of eye images based on images for comparison has always been a difficult problem in this field, and is also a major aspect limiting the application of eye analysis related instruments and equipment. How to extract traditional Chinese medicine choroid features from eye images is a common technical problem in the field of eye analysis instrument and equipment recognition technology. The current traditional Chinese medicine eye information analysis system recognition method is based on deep learning of all pixels of the full eye image to construct an AI-based analysis model. Although it is relatively easy to implement, the image has too much irrelevant element information, and direct machine learning without processing will greatly interfere with the final result and lead to poor accuracy. At the same time, due to the excessive number of input elements, a large number of samples are required, resulting in a long production cycle and poor stability. Due to the large number of samples, it takes a long time to re-learn the machine after modifying the samples each time, and it is impossible to respond quickly to problems, so it is difficult to apply in practice and cannot form a commercialized device.

[0003] Therefore, how to provide a method and device for recognizing human eye choroid features with good stability, high recognition accuracy, and fast response is a technical problem that needs to be solved in the field. SUMMARY

[0004] Based on the deficiencies of the prior art, the present application provides a white sclera choroid feature recognition method based on human eye white sclera zangfu partitioning. The above deficiencies can be improved.

[0005] As one aspect of the present application, the present application provides a white sclera choroid feature recognition method based on human eye white sclera zangfu partitioning, wherein the recognition method comprises,

[0006] obtaining a human eye image to be recognized;

[0007] inputting the human eye image to be recognized into a first mesh structure model to calculate the choroid morphology in the human eye image 、 color and direction feature point coordinate information;

[0008] sorting the feature point coordinate information to generate a feature point data unit, the sorting being based on the correspondence between the feature points and the choroid;

[0009] The human eye vascular features corresponding to the feature point data unit are calculated using the second mesh structure model. The human eye vascular features include vascular image information such as color, thickness, and spots.

[0010] Based on the feature point data units and human eye vein features, a first image is generated that includes at least human eye vein feature information.

[0011] Preferably, in the scleral vein feature recognition method provided by the first aspect of the present invention, the method further includes calculating the human eye vein type based on the human eye vein features, wherein the human eye vein type is information on vein morphology, vein color and special veins that correspond to human internal organs, and the human eye vein type includes partition information.

[0012] Preferably, the scleral vein feature recognition method provided in the first aspect of the present invention is characterized in that the human eye vein types include at least ten forms: thickened at the root, tortuous, extended, broken, forked, a raised line, a blurred patch, drooping, a dark circle, and penetrating the pupil.

[0013] And / or eight colors: bright red, purplish red, deep red, red with black, red with yellow, light yellow, gray, and dark gray.

[0014] And / or six special types of veins: straight, rootless, grid-like, black spots, yellow spots, and blue spots.

[0015] Preferably, in the scleral vein feature recognition method provided by the first aspect of the present invention, the second mesh structure model is trained using a support vector machine algorithm. The calculation of the human eye vein features corresponding to the feature point data unit through the second mesh structure model specifically includes: inputting the feature point data unit into the second mesh structure model and calculating the human eye vein features corresponding to the feature point data unit in the human eye image.

[0016] Preferably, the first aspect of the present invention provides a method for recognizing scleral vein features, wherein the first mesh structure model is trained by the following method:

[0017] Step a: Randomly generate one or more simple shapes, including quadrilaterals, triangles, line segments, cubes, and other similar shapes. Mark the vertices of the simple shapes as feature points. Use the coordinates of the simple shapes and their corresponding feature points as input to train an initial model using the MagicPoint algorithm.

[0018] Step b: Input the human eye image to be identified into the initial model trained by the MagicPoint algorithm, and perform a second round of training to obtain the advanced model.

[0019] Step c: Take the human eye image to be identified and the feature point coordinates calculated by the advanced model based on the human eye image to be identified as inputs, and use the MagicPoint algorithm to perform a third round of training to obtain a high-order model;

[0020] Step d: Using the human eye image to be identified as input, calculate the coordinates of the feature points corresponding to the human eye image to be identified using the higher-order model;

[0021] Step e: Manually check the accuracy of the feature points extracted in step d. If accurate, proceed to step g; otherwise, repeat step c to iterate the higher-order model.

[0022] Step g: Use the image of the human eye to be identified and the corresponding feature point coordinates as input, and train the final model using the SuperPoint algorithm.

[0023] Preferably, in the scleral vein feature recognition method provided by the first aspect of the present invention, the second mesh model is trained by the following steps:

[0024] Step A: Normalize each feature point data unit in the training set and convert it into a 4x32 two-dimensional matrix. Each column in the matrix corresponds to each line segment of the feature point data unit. The four data in each column are the vector angle, vector length, width, and color of a line segment of the feature point data unit. The 32 rows of data units can store a maximum of 32 line segment data. The extra data in the two-dimensional matrix that exceeds 32 line segments is discarded, and the line segments that are less than 32 are filled with 0.

[0025] Step B: Receive human eye vein features that are manually labeled as feature point data units;

[0026] Step C: Take the complex feature point data units that have been normalized in step A, and the human eye vein type of each vein feature data unit marked in step B as input, and train the model using the support vector machine algorithm.

[0027] Preferably, the first aspect of the present invention provides a method for recognizing scleral vein features, wherein acquiring the human eye image to be recognized specifically includes:

[0028] Import human eye images, which were captured by a camera.

[0029] Remove the background from the human eye image, retaining the eye image.

[0030] To remove highlighted areas caused by reflected light sources in images, a dilation algorithm is used to repair the removed areas.

[0031] The Canny edge detection algorithm, SOBEL edge detection algorithm, and dilation algorithm are used to identify and crop the image, retaining only the sclera region.

[0032] The human eye image that has been cropped to retain only the sclera region is used as the human eye image to be identified.

[0033] In another embodiment of the present invention, the present invention also provides a scleral vascular feature recognition device based on the human eye sclera viscera partitioning method, specifically comprising:

[0034] The acquisition unit acquires the image of the human eye to be identified;

[0035] The first computing unit inputs the human eye image to be identified into the first mesh structure model and calculates the vein morphology in the human eye image. 、 Color and direction of feature point coordinates;

[0036] The first generation unit classifies and sorts the feature point coordinate information to generate a feature point data unit, wherein the classification and sorting are based on the correspondence between the feature points and the network.

[0037] The second calculation unit calculates the human eye vein features corresponding to the feature point data unit through the second mesh structure model. The human eye vein features include blood vessel image information such as color, thickness, and spots.

[0038] The second generation unit generates a first image that includes at least human eye vein feature information based on the feature point data unit and human eye vein features.

[0039] In another embodiment of the present invention, a computer storage medium is provided, the computer storage medium storing a plurality of instructions adapted for loading and execution by a processor of the steps of the scleral vascular feature recognition method based on the human eye sclera organ partitioning method described above.

[0040] In another embodiment of the present invention, the present invention also provides an electronic device, comprising: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed as described above in the method for recognizing scleral vein features based on the human eye sclera viscera partitioning method.

[0041] The scleral vein feature recognition method based on the human eye sclera viscera partitioning method provided by this invention has the advantages of high recognition accuracy, small amount of adaptable data, rapid iteration in actual use, better versatility, and easy rapid production deployment. It provides a reliable and stable human eye scleral vein feature recognition method for the application of TCM eye analysis equipment.

[0042] This invention decomposes the identification of scleral vein features in the human eye into several steps: calculating feature point coordinates, classifying and sorting based on the correspondence between feature points and veins, calculating and classifying scleral vein features, and matching scleral vein features. This orderly classification of complex and ever-changing scleral vein feature information, using two separate models to complete different stages of calculation and identification, ensures data simplicity and achieves data processing stability. The modular approach also enables controllable samples, easy adaptive updates, rapid identification, and convenient production deployment. This invention solves the problems of low recognition accuracy, high sample quantity requirements, and slow sample training leading to long production and iteration cycles in current end-to-end deep learning for human eye image processing, providing a fundamental guarantee for accurate identification and widespread application in traditional Chinese medicine eye diagnosis.

[0043] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0045] Figure 1 This is a flowchart of a scleral vein feature recognition method based on the human eye's sclera organ partitioning method, as described in Embodiment 1 of the present invention.

[0046] Figure 2 This is a flowchart of the first mesh model training method in Embodiment 1 of the present invention.

[0047] Figure 3 This is a flowchart of the second mesh model training method in Embodiment 1 of the present invention.

[0048] Figure 4 This is a schematic diagram of the human eye scleral vein feature recognition device in Embodiment 2 of the present invention.

[0049] Figure 5 This is a schematic diagram of the electronic device structure in Embodiment 3 of the present invention. Detailed Implementation

[0050] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. Furthermore, the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0051] To address the problems of high data interference, poor accuracy, poor stability, inability to adapt and update in a timely manner according to actual use, slow response, and long production cycle in the field of human eye image recognition, this invention provides a method for recognizing the sclera vein features based on the human eye sclera organ partitioning method. It has the advantages of high recognition accuracy, small amount of adaptable data, rapid iteration in actual use, better versatility, and easy rapid production, providing a reliable and stable method for recognizing the sclera vein features of the human eye for the application of traditional Chinese medicine eye analysis equipment.

[0052] Specifically, the scleral vein feature recognition method based on the human eye sclera organ partitioning method in Embodiment 1 of the present invention specifically includes:

[0053] Step 101: Obtain the image of the human eye to be identified.

[0054] Specifically, the human eye image can be obtained by using an auxiliary device to fix and preset the human eye state of different people's eyes, and using a camera to take a picture of the target area of ​​the human eye to obtain the human eye image. In order to support subsequent image processing and algorithm model recognition, the preferred embodiment of the present invention is a camera that can clearly capture the details and colors of the eye's periphery, including the eye itself. A camera with corresponding high pixel and resolution, such as an SLR camera, a smartphone camera, or an industrial camera, can be selected.

[0055] In this embodiment of the invention, the human eye image to be identified is a human eye image obtained by taking a picture of the human eye with a camera and then processing the image data. Specifically, obtaining the human eye image to be identified includes:

[0056] Import human eye images, which were captured by a camera.

[0057] Remove the background from the human eye image, keeping only the eye image.

[0058] To remove highlighted areas caused by reflected illumination sources in images, an improved dilation algorithm is used to repair the removed areas.

[0059] The image is identified and cropped using the Canny edge detection algorithm, the SOBEL edge detection algorithm, and an improved dilation algorithm, retaining only the sclera region.

[0060] The human eye image that has been cropped to retain only the sclera region is used as the human eye image to be identified.

[0061] After processing the human eye image captured by the camera through the above steps, the eye area to be identified can be focused by cropping to the surrounding area. At the same time, the image influence factors caused by reflected light, background, edge blurring and other factors on the eye area to be identified can be eliminated, which can provide a high-quality human eye image for subsequent accurate processing, thereby improving the accuracy of the overall human eye vein feature recognition.

[0062] Step 102: Input the human eye image to be identified into the first mesh structure model, and calculate the vein morphology in the human eye image. 、 Color and direction of feature point coordinates.

[0063] Specifically, in this embodiment of the invention, feature points are points that can represent the shape characteristics of an image. For example, they can be selected from the four vertices of a square, the three vertices of a triangle, the eight vertices of a cube, the inflection points of a broken line, and the intersection points of intersecting lines. Connecting these points yields the outline of the graphic. Since the shape and direction of the human eye's visual pathways are related to the shape characteristics of an image, key points can be used for precise representation.

[0064] The feature point coordinate information calculated in Embodiment 1 of the present invention can be the coordinates of scattered feature points that do not correspond to the network state.

[0065] Specifically, as shown in the attached document Figure 2 In this embodiment of the invention, the first mesh structure model is trained using the following method:

[0066] Step a: Randomly generate one or more simple shapes, including quadrilaterals, triangles, line segments, cubes, and other similar shapes. Mark the vertices of the simple shapes as feature points. Use the coordinates of the simple shapes and their corresponding feature points as input to train an initial model using the MagicPoint algorithm.

[0067] Step b: Input the human eye image to be identified into the initial model trained by the MagicPoint algorithm, and perform a second round of training to obtain the advanced model.

[0068] Step c: Take the human eye image to be identified and the feature point coordinates calculated by the advanced model based on the human eye image to be identified as inputs, and use the MagicPoint algorithm to perform a third round of training to obtain a high-order model;

[0069] Step d: Using the human eye image to be identified as input, calculate the coordinates of the feature points corresponding to the human eye image to be identified using the higher-order model;

[0070] Step e: Manually check the accuracy of the feature points extracted in step d. If accurate, proceed to step g; otherwise, repeat step c to iterate the higher-order model.

[0071] Step g: Use the image of the human eye to be identified and the corresponding feature point coordinates as input, and train the final model using the SuperPoint algorithm.

[0072] This invention sets up a separate feature point coordinate information calculation model. In the first step, it quickly and accurately extracts feature points from the input human eye image information to be identified, and provides a foundation for subsequent fast and accurate processing in the form of feature point coordinate information. At the same time, the calculation process of feature points based on the human eye image to be identified has advantages such as simple data types and rules, small data volume, and fast and accurate calculation compared to directly calculating complete network features or network types.

[0073] Step 103: Classify and sort the feature point coordinate information to generate feature point data units. The classification and sorting are based on the correspondence between feature points and the network.

[0074] It should be noted that in this embodiment of the invention, the network refers to human tissue information that can transmit human biological signals, including blood vessels and nerve patterns in the sclera of the human eye, and is usually represented in the form of image information.

[0075] After calculating the feature point coordinates that characterize the shape and direction of the network, the resulting feature point coordinates are scattered feature point coordinates that are not associated with their respective network shapes and directions. The feature point coordinates calculated by the first network structure model are scattered feature points that are not associated with the network. However, these scattered feature points actually represent all the information contained in the network. Therefore, step 103, by classifying and sorting the scattered feature points according to their correspondence with each network, can achieve high efficiency in subsequent network feature calculations.

[0076] Therefore, in this embodiment of the invention, step 103 is used to further classify and sort based on the correspondence between feature points and the veins, filter out feature points that can be associated with the veins, and exclude feature points that are not associated with the actual location and shape of blood vessels in the image. These unassociated feature points are usually referred to as miscalculated feature points in the process. After removing miscalculated feature points, step 103 also sorts the feature points associated with the location and shape of the veins. The sorting is based on dividing the area where each vein is located in the image to be identified into data groups, and each data group corresponds to the location of each vein in the image to be identified.

[0077] In this embodiment of the invention, the classification and sorting of the feature point coordinate information in step 103 can be performed using existing technologies such as an adaptive pattern detector based on the topological structure of blood vessel patterns, or by machine learning trained on big data. The specific classification and sorting method can be selected and optimized by those skilled in the art based on the distribution patterns of blood vessels and nerves in the human eye and the calculated set of feature point coordinates.

[0078] Step 104: Calculate the human eye vein features corresponding to the feature point data unit using the second mesh structure model. The human eye vein features include vascular image information such as color, thickness, and spots. In this embodiment of the invention, step 104 involves training the second mesh structure model using a support vector machine algorithm. Calculating the human eye vein features corresponding to the feature point data unit using the second mesh structure model specifically includes: inputting the feature point data unit into the second mesh structure model and calculating the human eye vein features corresponding to the feature point data unit in the human eye image.

[0079] Specifically, as shown in the attached document Figure 2 As shown, the second mesh model in this invention is trained by the following steps:

[0080] Step A: Normalize each feature point data unit in the training set and convert it into a 4x32 two-dimensional matrix. Each column in the matrix corresponds to each line segment of the feature point data unit. The four data in each column are the vector angle, vector length, width, and color of a line segment of the feature point data unit. The 32 rows of data units can store a maximum of 32 line segment data. The extra data in the two-dimensional matrix that exceeds 32 line segments is discarded, and the line segments that are less than 32 are filled with 0.

[0081] Step B: Receive human eye vein features that are manually labeled as feature point data units;

[0082] Step C: Take the complex feature point data units that have been normalized in step A, and the human eye vein type of each vein feature data unit marked in step B as input, and train the model using the support vector machine algorithm.

[0083] The human eye vascular features described in this embodiment of the invention include at least vascular image information such as color, thickness, and spots. In addition to the above information, other vascular and nerve image information that can reflect the human eye vascular system may also be included in this invention.

[0084] In Embodiment 1 of this invention, through steps 102-104, the image of the human eye to be identified is processed by a model and algorithm to obtain human eye vascular features that include at least vascular and neural information. These features include the shape, direction, color, and thickness of the blood vessels in the eye, reflecting the distribution of blood vessels within the eye. Furthermore, to enable association and display with different eye states, the human eye vascular types described in this invention specifically include at least ten forms: thickened root, tortuous, extended, broken, bifurcated, a raised line, a blurred area, drooping, a black circle, and penetrating the pupil.

[0085] And / or eight colors: bright red, purplish red, deep red, red with black, red with yellow, light yellow, gray, and dark gray.

[0086] And / or six special types of veins: straight, rootless, grid-like, black spots, yellow spots, and blue spots.

[0087] Step 105: Based on the feature point data unit and the human eye vein features, generate a first image that includes at least human eye vein feature information.

[0088] Based on the theory of human eye condition classification in traditional Chinese medicine, and according to the correspondence between the characteristics and types of human eye meridians, the characteristics of human eye meridians are used to obtain the types of human eye meridians. This provides a guarantee for the automatic transformation of human eye image information after machine calculation into human eye condition information and provides automated human eye condition analysis. It has achieved a breakthrough in machine automation, high efficiency and accuracy, and plays a very important role in the application and promotion of eye diagnosis instruments.

[0089] It should be noted that the sclera-viscera regionalization method originates from the eye acupuncture theory of the renowned traditional Chinese medicine master, Professor Peng Jingshan. This method is part of a key national basic research and development program in traditional Chinese medicine (973 Program), project number 2007CB512707. It identifies eye information by dividing the left and right eyes into eight zones and thirteen acupoints, and then using these zones and the correspondence between blood vessel features and internal organs. In this embodiment of the invention, the sclera-viscera regionalization method is an application of the sclera-viscera regionalization method in computer image processing.

[0090] The method for recognizing the vascular features of the sclera of the human eye provided in Embodiment 1 of the present invention has the advantages of high recognition accuracy, adaptability to large amounts of data and rapid iteration in actual use, better versatility and easy rapid production, and provides a reliable and stable method for recognizing the vascular features of the sclera of the human eye for application in the field of traditional Chinese medicine ophthalmology.

[0091] This invention decomposes the identification of scleral vein features in the human eye into several steps: calculating feature point coordinates, classifying and sorting based on the correspondence between feature points and veins, calculating and classifying scleral vein features, and matching scleral vein features. This method orderly classifies complex and ever-changing information related to scleral vein features. By using two separate models to complete the calculation and identification at different stages, the invention ensures the accuracy of the data and achieves stability in data processing. The modular approach also enables controllable samples, easy adaptive updates, rapid identification, and convenient production deployment. This invention solves the problems of low recognition accuracy, high sample quantity requirements, and slow sample training leading to long production and iteration cycles in current end-to-end deep learning for human eye image processing, providing a fundamental guarantee for accurate identification and widespread application in traditional Chinese medicine eye diagnosis.

[0092] As attached Figure 4 As shown, the second embodiment of the present invention provides a device for identifying the type of scleral veins in the human eye, specifically including an acquisition unit 401 for acquiring an image of the human eye to be identified;

[0093] The first computing unit 403 inputs the human eye image to be identified into the first mesh structure model and calculates the vein morphology in the human eye image. 、 Color and direction of feature point coordinates;

[0094] The first generation unit 405 classifies and sorts the feature point coordinate information to generate a feature point data unit, wherein the classification and sorting are based on the correspondence between the feature points and the network.

[0095] The second calculation unit 407 calculates the human eye vein features corresponding to the feature point data unit through the second mesh structure model. The human eye vein features include blood vessel image information such as color, thickness, and spots.

[0096] The second generation unit 409 generates a first image that includes at least human eye vein feature information based on the feature point data unit and human eye vein features.

[0097] In this embodiment of the invention, the human eye image to be identified is a human eye image obtained by taking a picture of the human eye with a camera and then processing the image data. Specifically, obtaining the human eye image to be identified includes:

[0098] Import human eye images, which were captured by a camera.

[0099] Remove the background from the human eye image, keeping only the eye image.

[0100] To remove highlighted areas caused by reflected illumination sources in images, an improved dilation algorithm is used to repair the removed areas.

[0101] The image is identified and cropped using the Canny edge detection algorithm, the SOBEL edge detection algorithm, and an improved dilation algorithm, retaining only the sclera region.

[0102] The human eye image that has been cropped to retain only the sclera region is used as the human eye image to be identified.

[0103] After processing the human eye image captured by the camera through the above steps, the peripheral area can be cropped to focus on the eye area to be identified. At the same time, the image influence factors caused by reflected light, background, edge blurring and other factors on the eye area to be identified are eliminated, which can provide a high-quality human eye image for subsequent accurate processing, thereby improving the accuracy of the overall human eye sclera feature recognition.

[0104] The composition and training method of the first and second mesh structure models in this embodiment of the invention are the contents that have been fully disclosed in Embodiment 1, and will not be described again here.

[0105] Appendix Figure 4 The device for recognizing the scleral vein features of the human eye shown in Embodiment 2 of the present invention corresponds to the method for recognizing the scleral vein features of the human eye in Embodiment 1. By decomposing the recognition of the scleral vein features of the human eye into the following steps: calculation of feature point coordinate information, classification and sorting based on the correspondence between feature points and veins, calculation of the scleral vein features of the human eye, and classification and matching of the scleral vein features of the human eye, the complex and ever-changing information related to the scleral vein features of the human eye is orderly classified. By using two models to complete the calculation and recognition at different stages, the simplicity of the data is ensured, thereby achieving the stability of data processing. The modular approach also enables controllable samples, easy adaptive updates, rapid recognition, and convenient production. It solves the problems of low recognition accuracy, high sample quantity requirements, and slow sample training leading to long production and iteration cycles in current end-to-end deep learning in human eye image processing, and provides innovative support for the accurate recognition and widespread application of traditional Chinese medicine eye diagnosis.

[0106] Appendix Figure 5 An example is a schematic diagram of the electronic device structure in Embodiment 3 of the present invention, as shown below. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a scleral vein feature recognition method based on the human eye sclera organ partitioning method of the present invention, the method including: acquiring an image of the human eye to be identified;

[0107] The human eye image to be identified is input into a first mesh structure model to calculate the vein morphology in the human eye image. 、 Color and direction of feature point coordinates;

[0108] The feature point coordinate information is classified and sorted to generate feature point data units, and the classification and sorting are based on the correspondence between feature points and the network.

[0109] The human eye vascular features corresponding to the feature point data unit are calculated using the second mesh structure model. The human eye vascular features include vascular image information such as color, thickness, and spots.

[0110] Based on the feature point data units and human eye vein features, a first image is generated that includes at least human eye vein feature information.

[0111] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0112] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to execute the sclera vein feature recognition method based on the human eye sclera organ partitioning method provided in the above embodiments, the method including: acquiring a human eye image to be identified;

[0113] The human eye image to be identified is input into the first mesh structure model to calculate the coordinate information of feature points in the human eye image, including the morphology, color and direction of the veins.

[0114] The feature point coordinate information is classified and sorted to generate feature point data units, and the classification and sorting are based on the correspondence between feature points and the network.

[0115] The human eye vascular features corresponding to the feature point data unit are calculated using the second mesh structure model. The human eye vascular features include vascular image information such as color, thickness, and spots.

[0116] Based on the feature point data units and human eye vein features, a first image is generated that includes at least human eye vein feature information.

[0117] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the above-mentioned method for recognizing scleral vein features based on the human eye sclera organ partitioning method, the method comprising: acquiring an image of a human eye to be recognized;

[0118] The human eye image to be identified is input into a first mesh structure model to calculate the vein morphology in the human eye image. 、 Color and direction of feature point coordinates;

[0119] The feature point coordinate information is classified and sorted to generate feature point data units, and the classification and sorting are based on the correspondence between feature points and the network.

[0120] The human eye vascular features corresponding to the feature point data unit are calculated using the second mesh structure model. The human eye vascular features include vascular image information such as color, thickness, and spots.

[0121] Based on the feature point data units and human eye vein features, a first image is generated that includes at least human eye vein feature information.

[0122] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0123] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for recognizing scleral vein features based on the human eye's sclera organ zonal division method, characterized in that, The identification method includes, Obtain the image of the human eye to be identified; The human eye image to be identified is input into a first mesh structure model to calculate the coordinate information of feature points in the image, including the morphology, color, and direction of the veins. The first mesh structure model is trained using the following method: Step a: Randomly generate one or more simple shapes, including quadrilaterals, triangles, line segments, and cubes. Mark the vertices of the simple shapes as feature points. Use the simple shapes and their corresponding feature point coordinates as input to train an initial model using the MagicPoint algorithm. Step b: Input the human eye image to be identified into the initial model trained by the MagicPoint algorithm for a second round of training to obtain an advanced model. c. Using the image of the human eye to be identified and the coordinates of the feature points calculated by the advanced model based on the image of the human eye to be identified as input, perform a third round of training using the MagicPoint algorithm to obtain a high-order model; d. Using the image of the human eye to be identified as input, calculate the coordinates of the feature points corresponding to the image of the human eye to be identified using the high-order model; e. Manually check the accuracy of the feature points extracted in step d. If accurate, proceed to g; if inaccurate, repeat step c to iterate the high-order model; g. Using the image of the human eye to be identified and the corresponding coordinates of the feature points as input, train the final model using the SuperPoint algorithm. The feature point coordinate information is classified and sorted to generate feature point data units, and the classification and sorting are based on the correspondence between feature points and the network. The human eye vein features corresponding to the feature point data unit are calculated using a second mesh structure model. The human eye vein features include vascular image information such as color, thickness, and spots. The second mesh structure model is trained using a support vector machine algorithm. Calculating the human eye vein features corresponding to the feature point data unit using the second mesh structure model specifically includes: inputting the feature point data unit into the second mesh structure model and calculating the human eye vein features corresponding to the feature point data unit in the human eye image. Based on the feature point data units and human eye vein features, a first image is generated that includes at least human eye vein feature information.

2. The scleral vein feature recognition method as described in claim 1, characterized in that, The method further includes calculating the human eye vein type based on the human eye vein characteristics. The human eye vein type is information on vein morphology, vein color, and special veins that correspond to the human internal organs. The human eye vein type includes zoning information.

3. The scleral vein feature recognition method as described in claim 2, characterized in that, The types of human eye veins include at least ten forms: thick root, tortuous, extended, broken, forked, raised, blurred, drooping, black circle, and penetrating the pupil; and / or eight colors: bright red, purplish red, dark red, red with black, red with yellow, light yellow, gray, and dark gray; and / or six special veins: straight, weak root, grid-like, black spot, yellow spot, and blue spot.

4. The scleral vein feature recognition method as described in claim 1, characterized in that, The second mesh structure model is trained using the following steps: Step A: Normalize each feature point data unit in the training set and convert it into a 4x32 two-dimensional matrix. Each column in the matrix corresponds to each line segment of the feature point data unit. The four data in each column are the vector angle, vector length, width, and color of a line segment of the feature point data unit. The extra data in the two-dimensional matrix that exceeds 32 line segments are discarded, and the line segments that are less than 32 are filled with 0. Step B: Receive human eye vein features that are manually labeled as feature point data units; Step C: Take the complex feature point data units that have been normalized in step A, and the human eye vein type of each vein feature data unit marked in step B as input, and train the model using the support vector machine algorithm.

5. The scleral vein feature recognition method as described in claim 1, characterized in that, The acquisition of the human eye image to be identified specifically includes: Import human eye images, which were captured by a camera. Remove the background from the human eye image, retaining the eye image. To remove highlighted areas caused by reflected light sources in images, a dilation algorithm is used to repair the removed areas. The Canny edge detection algorithm, SOBEL edge detection algorithm, and dilation algorithm are used to identify and crop the image, retaining only the sclera region. The human eye image that has been cropped to retain only the sclera region is used as the human eye image to be identified.

6. A device for recognizing scleral vein features based on the human eye's sclera organ zonal division method, characterized in that, Specifically, it includes: The acquisition unit acquires the image of the human eye to be identified; The first computing unit inputs the human eye image to be identified into a first mesh structure model and calculates the feature point coordinates information of the vein shape, color, and direction in the human eye image. The first mesh structure model is trained by the following method: Step a: Randomly generate one or more simple shapes, including quadrilaterals, triangles, line segments, cubes, etc., mark the vertices of the simple shapes as feature points, and use the simple shapes and the corresponding feature point coordinates as input to train the initial model using the MagicPoint algorithm. Step b: Input the human eye image to be identified into the initial model trained by the MagicPoint algorithm for a second round of training to obtain an advanced model. Step c: Using the image of the human eye to be identified and the feature point coordinates calculated by the advanced model based on the image of the human eye to be identified as input, perform a third round of training using the MagicPoint algorithm to obtain a high-order model; Step d: Using the image of the human eye to be identified as input, calculate the feature point coordinates corresponding to the image of the human eye to be identified using the high-order model; Step e: Manually check the accuracy of the feature points extracted in step d. If accurate, proceed to step g; if inaccurate, repeat step c to iterate the high-order model; Step g: Using the image of the human eye to be identified and the corresponding feature point coordinates as input, train the final model using the SuperPoint algorithm. The first generation unit classifies and sorts the feature point coordinate information to generate a feature point data unit, wherein the classification and sorting are based on the correspondence between the feature points and the network. The second calculation unit calculates the human eye vein features corresponding to the feature point data unit through a second mesh structure model. The human eye vein features include blood vessel image information such as color, thickness, and spots. The second mesh structure model is trained using a support vector machine algorithm. Calculating the human eye vein features corresponding to the feature point data unit through the second mesh structure model specifically includes: inputting the feature point data unit into the second mesh structure model and calculating the human eye vein features corresponding to the feature point data unit in the human eye image. The second generation unit generates a first image that includes at least human eye vein feature information based on the feature point data unit and human eye vein features.

7. A computer storage medium storing a plurality of instructions, the instructions being adapted to be loaded by a processor and executed by the steps of the sclera vein feature recognition method based on the human eye sclera organ partitioning method as claimed in any one of claims 1-5.

8. An electronic device, comprising: A processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed the steps of the scleral vascular feature recognition method based on the human eye sclera viscera partitioning method as claimed in any one of claims 1-5.

Images