Iris image correction method, device, equipment and storage medium

Abstract

The embodiments of the present application provide an iris image correction method, device, equipment and storage medium, which relate to the field of iris recognition technology. The method includes acquiring iris data, the iris data including multiple groups of iris images; preprocessing the iris data, and dividing the preprocessed iris data into a training data set and a verification data set; constructing an iris image correction deep neural network model, and training the model according to the training data set, and verifying it through the verification data set to obtain a trained model; inputting the iris image to be corrected into the trained model for forward reasoning to obtain a corrected iris image; performing linear normalization processing on the corrected iris image to obtain a corrected normalized image. In this way, an iris image of any shape can be corrected into a concentric circle shape with a uniform scale, providing a high-quality foundation for iris normalization calculation.

Description

Technical Field

[0001] The present application relates to the technical field of iris recognition, and in particular to an iris image correction method, device, equipment and storage medium. Background Art

[0002] Iris recognition technology is a technology that uses the iris in the human eye to authenticate identity. It is a type of human biometric technology. Iris recognition has the characteristics of uniqueness, stability, non-contact, and high security. It is recognized as the most accurate and convenient biometric technology and has been widely used in finance, security, checkpoints, access control, insurance and other scenarios that require accurate identity authentication.

[0003] One of the key stages of classic iris recognition is the normalization process, which maps the annular iris area to a dimensionless pseudo-polar coordinate system. This process produces a rectangular structure to compensate for differences in image size ratios and changes in pupil scaling. The normalization process usually uses the isotropic rubber sheet model proposed by Daugman, with linear sampling in the radial and angular directions. This normalization method will bring two problems: First, the iris image is essentially a projection of a three-dimensional iris on a two-dimensional image plane. Affected by the user's cooperation, the actual iris image is usually multi-pose (for example, different parameters, non-concentric elliptical projections). Linear normalization of this iris image will cause distortion and dislocation of the normalized image; second, the texture change caused by pupil scaling is itself nonlinear, and the use of linear normalization will cause overall deformation of the normalized image. Summary of the invention

[0004] The present application provides an iris image correction method, apparatus, device and storage medium, which can correct an iris image of any shape into a concentric circle shape with a uniform scale, as a basis for iris normalization calculation.

[0005] According to a first aspect of the present application, a method for correcting an iris image is provided, the method comprising:

[0006] Acquire iris data, wherein the iris data includes multiple groups of iris images, and each group of iris images has unique identification information;

[0007] Preprocessing the iris data, and dividing the preprocessed iris data into a training data set and a verification data set;

[0008] Constructing an iris image correction deep neural network model, and training the iris image correction deep neural network model according to the training data set, and verifying it through the verification data set to obtain a trained iris image correction deep neural network model;

[0009] Inputting the iris image to be corrected into the trained iris image correction deep neural network model for forward reasoning to obtain a corrected iris image;

[0010] The corrected iris image is subjected to linear normalization processing to obtain a corrected normalized image.

[0011] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the preprocessing of the iris data comprises:

[0012] Selecting an input image and a target image from the iris data;

[0013] Preprocessing the input image to obtain a preprocessed input image, an input image mask, and a target mask of an output image;

[0014] The target image is preprocessed to obtain a preprocessed target image and a target image mask.

[0015] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein a target image is selected from the iris data, comprising:

[0016] Image analysis technology was used to evaluate the concentricity and clarity of the iris images in each group;

[0017] The clear image in each group of iris images whose concentricity deviation from the ideal concentric circle state is the smallest and whose clarity reaches the set threshold is selected as the target image.

[0018] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the input image is preprocessed to obtain a preprocessed input image and an input image mask, and a target mask of an output image, including:

[0019] Positioning and segmenting the input image, identifying a valid iris area and an invalid iris area, and marking the invalid iris area as 0;

[0020] Determine the center of the effective iris area, and cut the effective iris area based on the center of the effective iris area to obtain a standard cut area that can include all effective iris information;

[0021] Adjusting the size of the standard cutting area and performing data standardization processing on all valid iris information in the standard cutting area to obtain a preprocessed input image;

[0022] Obtaining the size information of an input image and creating a blank image having the same size information as the input image;

[0023] Marking the valid iris region and the invalid iris region of the input image in the blank image to obtain an input image mask;

[0024] The input image mask is subjected to concentric ring processing to obtain a target mask of the output image.

[0025] According to the above aspects and any possible implementation manner, an implementation manner is further provided, in which the target image is preprocessed to obtain a preprocessed target image and a target image mask, including:

[0026] Positioning and segmenting the target image, identifying a valid iris area and an invalid iris area, and marking the invalid iris area as 0;

[0027] Determine the center of the effective iris area, and crop the effective iris area based on the center of the effective iris area to obtain a standard cropped image, wherein the standard cropped image includes all effective iris information;

[0028] Performing concentric circularization processing on the standard cropped image to obtain a concentric circularized image, wherein the concentric circularized image has the same geometric shape as the standard cropped image;

[0029] The size of the concentrically annularized image is adjusted and all valid iris information in the concentrically annularized image is subjected to data standardization processing to obtain a preprocessed target image;

[0030] Obtaining size information of a target image, and creating a blank image having the same size information as the target image;

[0031] The valid iris region and the invalid iris region of the target image are marked in the blank image to obtain a target image mask.

[0032] According to the above aspects and any possible implementation, an implementation is further provided, wherein the step of constructing a deep neural network model for iris image correction includes:

[0033] The U-shaped network structure is used to design the deep neural network model structure for iris image correction, wherein the U-shaped network structure includes an input processing module, an output processing module, a downsampling module, an upsampling module, and a jumper module, and the overall structure is a U-shaped symmetrical structure;

[0034] Determining the training loss of the iris image correction deep neural network model structure by combining multiple losses with different weights;

[0035] An iris image correction deep neural network model is constructed according to the iris image correction deep neural network model structure and the iris image correction deep neural network model structure training loss.

[0036] According to the above aspects and any possible implementation, an implementation is further provided, wherein the loss of the iris image correction deep neural network model structure training is determined by combining multiple losses with different weights, including:

[0037] Determine a joint loss function according to the loss values ​​calculated by the smoothL1 function, the L1loss function and the SSIM function and their corresponding loss calculation weights, and use the joint loss function as the training loss of the deep neural network model structure for iris image correction;

[0038] Among them, the smoothL1 function is used to calculate the difference loss value between the output image and the target image; the L1loss function is used to calculate whether the mask of the actual output image is consistent with the mask target of the output image; the SSIM function is used to calculate the structural difference loss value between the output image and the target image.

[0039] According to the above aspects and any possible implementation, an implementation is further provided, wherein the joint loss function is:

[0040] ;

[0041] in, represents the total joint loss value, Represents each loss item, For the The calculation weight of the loss, For the The calculated value of the loss.

[0042] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the training of the iris image correction deep neural network model according to the training data set comprises:

[0043] Step S91: input the training data set into the iris image correction deep neural network model for forward reasoning to obtain the current reasoning result;

[0044] Step S92: Calculate the loss between the current inference result and the target mask of the output image through a joint loss function to obtain a current joint loss;

[0045] Step S93: The current joint loss is transmitted back to the iris image segmentation deep neural network model through the reverse gradient propagation mechanism, and the current network weight is adjusted in combination with the learning rate;

[0046] The steps S91 to S93 are executed cyclically until the training reaches a specified preset number of cycles, and the training ends.

[0047] According to a second aspect of the present application, an iris image correction device is provided. The device comprises:

[0048] An iris data acquisition module, used to acquire iris data, wherein the iris data includes multiple groups of iris images, and each group of iris images has unique identification information;

[0049] An iris data preprocessing module, used to preprocess the iris data and divide the preprocessed iris data into a training data set and a verification data set;

[0050] A deep neural network model construction module, used to construct an iris image correction deep neural network model, and train the iris image correction deep neural network model according to the training data set, and verify it through the verification data set to obtain a trained iris image correction deep neural network model;

[0051] An iris image correction module, used for inputting the iris image to be corrected into the trained iris image correction deep neural network model for forward reasoning to obtain a corrected iris image;

[0052] The linear normalization processing module is used to perform linear normalization processing on the corrected iris image to obtain a corrected normalized image.

[0053] According to a third aspect of the present application, an electronic device is provided. The electronic device comprises: a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, the method described above is implemented.

[0054] According to a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method according to the first aspect of the present application is implemented.

[0055] Compared with the prior art, the present invention achieves the following beneficial effects:

[0056] The present invention can correct iris images of any shape into concentric circles of uniform scale, providing a better foundation for iris normalization calculation, and effectively solving the problems of image distortion, dislocation and deformation caused by traditional iris normalization methods when processing multi-pose iris images and nonlinear pupil scaling, significantly improving the accuracy and reliability of iris recognition, and enhancing the application effect of iris recognition technology in many precise identity authentication scenarios such as finance, security, checkpoints, access control, insurance, etc.

[0057] It should be understood that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of the present application, nor are they intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present application. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0059] Figure 1 A flow chart of an iris image correction method according to an embodiment of the present application is shown;

[0060] Figure 2 A schematic diagram of iris image preprocessing according to an embodiment of the present application is shown;

[0061] Figure 3 A structural diagram of a deep neural network model according to an embodiment of the present application is shown;

[0062] Figure 4 A ConvBlock structure diagram according to an embodiment of the present application is shown;

[0063] Figure 5 A DownBlock structure diagram according to an embodiment of the present application is shown;

[0064] Figure 6 shows a structural diagram of UpBlock according to an embodiment of the present application;

[0065] Figure 7 shows a BaseBlock structure diagram according to an embodiment of the present application;

[0066] Figure 8 A block diagram of an iris image correction device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0068] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0069] Embodiment 1: Figure 1 A flow chart of a method 100 for processing an iris image correction is shown. Figure 1 As shown, the method comprises the following steps:

[0070] S110, acquiring iris data, wherein the iris data includes multiple groups of iris images, and each group of iris images has unique identification information.

[0071] In some embodiments, in order to ensure that the model can adapt to a variety of actual scenarios and enhance generalization capabilities, it is necessary to ensure that the number of iris images is sufficient and diverse. The diversity needs to cover different populations, different collection environments and different occlusion conditions including but not limited to eyelids, eyelashes, hair, external reflections, different wearing items (such as frame glasses, contact lenses, etc.), etc.

[0072] Among them, the unique identification information of each group of iris images can be ID information, that is, each ID information has a corresponding group of iris images. Using ID information as the unique identification information of each group of iris images can clearly classify and organize a large amount of iris image data. In the data preprocessing stage, it is convenient to quickly locate and filter out iris images belonging to the same group based on ID, ensure the correct pairing of input images with target images and masks, and improve the accuracy and efficiency of data processing. In addition, during the model training process, ID information helps the model to better learn and distinguish the differences in iris features of different individuals. The model can perform targeted feature extraction and pattern recognition for the image data under each ID, thereby improving the generalization ability of different individual irises. This enables the trained model to more accurately correct and identify iris images of different IDs in actual applications, improve the reliability and stability of the entire iris recognition system, and effectively meet the needs of high precision and high security for identity authentication in the fields of finance and security.

[0073] S120, preprocessing the iris data, and dividing the preprocessed iris data into a training data set and a verification data set.

[0074] In some embodiments, the iris data is preprocessed. First, an input image and a target image need to be selected from the iris data. Further, the input image and the target image need to be preprocessed respectively. Specifically, when selecting the target image, the image analysis technology needs to be used to evaluate the iris images in each group respectively. The evaluation dimensions include but are not limited to the concentricity, clarity, roundness, etc. of the iris image. Concentricity, clarity and roundness are all important indicators for measuring the quality of iris images. By evaluating the concentricity, clarity and roundness of the iris image, a clear image with the smallest deviation between the concentricity and the ideal concentric circle state, the clarity reaching a set threshold and the roundness reaching a set threshold in each group of iris images is selected as the target image. This can reduce training errors and ensure the final correction effect and recognition accuracy.

[0075] Furthermore, the input image is preprocessed, including:

[0076] (121) Positioning and segmenting the input image, identifying the valid iris area and the invalid iris area, retaining only the valid iris area, and marking the invalid iris area as 0;

[0077] (122) The effective iris area is cropped based on the center of the effective iris area to obtain a standard cropped area that can include all effective iris information. In this embodiment, the standard cropped area is a square.

[0078] (123) Adjust the size of the standard cropping area to a size suitable for model input, such as 128*128 pixels;

[0079] (124) Performing data standardization processing on all valid iris information in the standard cropping area to obtain a preprocessed input image. Specifically, the data standardization processing method is:

[0080] ;

[0081] in, is the pixel value of the image after normalization. is the image pixel value before normalization, is the statistical average of the pixel values ​​of the training data set, is the standard deviation of the pixel values ​​in the training dataset.

[0082] At the same time, it is also necessary to generate a mask of the input image, that is, create a blank image with the same size information as the input image, and mark the valid iris area and invalid iris area of ​​the input image in the blank image. For example, 0 represents the invalid iris area and 1 represents the valid iris area, and the input image mask is obtained. Further, the input image mask is concentrically annularized to obtain the target mask of the output image.

[0083] It is worth noting that the target mask of the output image is obtained by concentric ring processing of the input image mask, which provides a clear output expectation for the model and stipulates the ideal state that the effective area of ​​the iris image should present after correction, that is, in concentric rings of fixed size. This enables the model to adjust parameters towards this specific goal during training. For example, during the forward reasoning and back propagation process of the U-shaped network structure, the model will continuously optimize its weights according to the difference with the target mask to reduce the error between the output image and the target image at the mask level, thereby guiding the model to learn the correct image correction transformation method. In addition, the target mask processed by concentric ring processing ensures that the output image of all training data is consistent in format. Regardless of the initial form of the input image, the model aims to correct it to the concentric ring structure defined by the target mask. This allows different iris images to maintain a relatively uniform position and shape of their effective areas after being processed by the model, which is conducive to subsequent linear normalization operations.

[0084] Furthermore, the target image is preprocessed, including:

[0085] (125) Positioning and segmenting the target image, identifying the valid iris area and the invalid iris area, retaining only the valid iris area, and marking the invalid iris area as 0;

[0086] (126) cropping the effective iris area based on the center of the effective iris area to obtain a standard cropping area that can include all effective iris information. In this embodiment, the standard cropping area is a square;

[0087] (127) performing concentric annularization processing on the standard cropped image to obtain a concentric annularized image, ensuring that all valid iris information is only in concentric rings of a fixed size, wherein the concentric annularized image has the same geometric shape as the standard cropped image. In this embodiment, the processed image is still a square;

[0088] (128) Adjusting the size of the concentrically circularized image to a size suitable for model input, such as 128*128 pixels;

[0089] (129) Performing data standardization processing on all valid iris information in the standard cropping area to obtain a preprocessed input image. Specifically, the data standardization processing method is:

[0090] ;

[0091] in, is the pixel value of the image after normalization. is the image pixel value before normalization, is the statistical average of the pixel values ​​of the training data set, is the standard deviation of the pixel values ​​in the training dataset.

[0092] At the same time, it is also necessary to generate a mask of the target image, that is, create a blank image with the same size information as the target image, and mark the valid iris area and invalid iris area of ​​the input image in the blank image. For example, use 0 to represent the invalid iris area and 1 to represent the valid iris area to obtain the target image mask.

[0093] It is worth mentioning that the mask of the generated target image provides a key comparison benchmark for evaluating the accuracy of the model output. The existence of the target image mask ensures the consistency of all images in the definition of the effective area. Different iris images may have various differences, but through the unified mask standard, they can follow the same rules when entering the model training and processing flow, so that the model can process different input images in a consistent manner, avoid processing deviations caused by differences in the definition of the effective area of ​​the image, improve the generalization ability and stability of the model, and ensure the reliability and repeatability of the final correction results.

[0094] The schematic diagram of preprocessing the iris image in this embodiment is as follows: Figure 2 As shown, from left to right are the preprocessed input image, the preprocessed target image and target image mask, and the target mask of the output image (in order to show the concentric ring effect, the invalid iris area is marked in gray within the ring, and the valid iris area is marked in white).

[0095] In some embodiments, the preprocessed input image and target image may be divided into a training data set and a verification data set in a certain ratio, for example, the ratio of the training data set to the verification data set is 8:2.

[0096] S130, constructing an iris image correction deep neural network model, and training the iris image correction deep neural network model according to the training data set, and verifying it through the verification data set to obtain a trained iris image correction deep neural network model.

[0097] In some embodiments, constructing the iris image correction deep neural network model includes designing a network structure of the iris image correction deep neural network model and a training loss of the iris image correction deep neural network model structure. Specifically, designing the network structure of the iris image correction deep neural network model includes the following steps:

[0098] (131) The model adopts a U-shaped network structure, including an input processing module, an output processing module, a downsampling module, an upsampling module, a jumper module, and the entire structure is a U-shaped symmetrical structure. The model structure in this embodiment is as follows Figure 3As shown in the figure, it includes 1 input processing module, 1 output processing module, 4 down-sampling modules, 4 up-sampling modules, and 3 jumper modules. The overall structure is U-shaped and symmetrical. The detailed information of the model structure is shown in the attached figure. Figure 4 -Attached Figure 7 As shown, the following Table 1 shows the detailed network structure:

[0099] Table 1:

[0100] ;

[0101] (132) The input processing module is responsible for rapidly increasing the dimension of the input data and adapting it to the subsequent downsampling module input.

[0102] (133) The downsampling module continues to increase the data dimension while extracting deep features.

[0103] (134) The upsampling module gradually reduces the data dimension and restores the image information.

[0104] (135) The jumper module converts the low-level information of the symmetrical position and realizes the conversion and restoration of image information together with the corresponding upsampling module.

[0105] (136) The output processing module aggregates the output image information and generates converted image information of the same size as the input image.

[0106] The design of the training loss of the deep neural network model structure for iris image segmentation includes: determining a joint loss function according to the loss values ​​calculated by the smoothL1 function, the L1loss function and the SSIM function and their corresponding loss calculation weights, and using the joint loss function as the training loss of the deep neural network model structure for iris image correction;

[0107] Among them, the smoothL1 function is used to calculate the difference loss value between the output image and the target image; the L1loss function is used to calculate whether the mask of the actual output image is consistent with the mask target of the output image; the SSIM function is used to calculate the structural difference loss value between the output image and the target image.

[0108] Among them, the smoothL1 function is used to calculate the difference loss value between the output image and the target image , the formula is:

[0109] ;

[0110] in, The numerical difference between the output image and the target image.

[0111] The L1loss function is used to calculate whether the mask of the actual output image is consistent with the mask target of the output image:

[0112] ;

[0113] in, is the numerical difference between the actual mask image of the output image and the target mask image of the output image.

[0114] The SSIM function is used to calculate the structural difference loss value between the output image and the target image , the formula is:

[0115] ;

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] represents the rectified image, represents the target image, is the pixel mean of the rectified image, is the pixel mean of the target image; For the rectified image The standard deviation of is the covariance between the rectified image and the target image, Represents the pixels in the image, Represents the total number of pixels in the image. Represents the first pixels, Indicates the target image pixels, is a constant used to improve calculation stability and prevent the denominator from being too small. =0.01, =0.02.

[0121] The loss values ​​calculated by the above smoothL1 function, L1loss function and SSIM function are superimposed according to different weights, and the calculation method is:

[0122] ;

[0123] in, represents the total joint loss value, Represents each loss item, with values ​​of 1, 2, and 3. For the The calculation weight of the loss, For the The calculated value of the loss term, in this embodiment, .

[0124] In some embodiments, training the iris image segmentation deep neural network model according to the training data set comprises the following steps:

[0125] Step S91: input the training data set into the iris image correction deep neural network model for forward reasoning to obtain the current reasoning result;

[0126] Step S92: Calculate the loss between the current inference result and the target mask of the output image through a joint loss function to obtain a current joint loss;

[0127] Step S93: The current joint loss is transmitted back to the iris image segmentation deep neural network model through the reverse gradient propagation mechanism, and the current network weight is adjusted in combination with the learning rate;

[0128] The steps S91 to S93 are executed repeatedly until the training reaches the specified preset number of cycles, and the training ends. Meanwhile, during the training process, the validation data set is used to perform network model validation and parameter screening, and the network model parameter screening condition is that the index of the validation data set is the best result of the current training and meets expectations.

[0129] S140, inputting the iris image to be corrected into the trained iris image correction deep neural network model for forward reasoning to obtain a corrected iris image.

[0130] In some embodiments, the iris image to be segmented is first preprocessed in a manner that is exactly the same as the method used in the above step S120. The preprocessed iris image is input into the trained iris image segmentation deep neural network model for forward reasoning to obtain a corrected iris image.

[0131] S150, performing linear normalization processing on the corrected iris image to obtain a corrected normalized image.

[0132] In some embodiments, the iris image that is corrected and is in a concentric ring of a preset size can be converted from its original image coordinate system to a dimensionless pseudo-polar coordinate system. In this process, linear sampling is mainly performed along the radial and angular directions. Through such processing, the difference in image size ratio and the changes caused by pupil scaling can be effectively compensated, so that the iris images of different individuals and under different acquisition conditions are in a unified standard framework after correction and linear normalization, which facilitates subsequent accurate iris feature extraction, matching and recognition operations, improves the stability and accuracy of the entire iris recognition system, and ensures that it can function reliably in application scenarios such as finance and security that require extremely high accuracy in identity authentication.

[0133] According to the embodiments of the present disclosure, the following beneficial effects are achieved:

[0134] (1) The present invention can correct iris images of any shape into concentric circles of uniform scale, providing a better foundation for iris normalization calculation, and effectively solving the problems of image distortion, dislocation and deformation caused by traditional iris normalization methods when processing multi-pose iris images and nonlinear pupil scaling, significantly improving the accuracy and reliability of iris recognition, and enhancing the application effect of iris recognition technology in many precise identity authentication scenarios such as finance, security, checkpoints, access control, insurance, etc.

[0135] (2) The present invention performs comprehensive and detailed preprocessing operations on the input image, the target mask of the output image, and the target image and mask, and performs concentric ring processing on the target image, the target mask, and the target mask of the output image to ensure that the output results of the model meet expectations, provide high-quality and representative targets for model learning, improve the overall data quality, and lay a good foundation for model training.

[0136] (3) The unique U-shaped network structure is adopted, and each module cooperates with each other. The input processing module increases the dimension to adapt to the input, the downsampling module extracts deep features, the upsampling module restores image information, the jumper module assists in information conversion and recovery, and the output processing module generates an output image of appropriate size. This structural design enables the model to have powerful iris image correction capabilities, can efficiently process image data, and achieve complex image correction tasks.

[0137] (4) The joint loss function combines smoothL1, L1loss and SSIM losses, and superimposes them by reasonably setting different weights. It can comprehensively measure the errors between the model output and the target data in terms of numerical difference, mask consistency and structural difference, etc., and provides precise optimization direction for model training, prompting the model to continuously adjust its own parameters during the training process, improve the accuracy and stability of the output results, and thus improve the performance of the entire correction method.

[0138] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0139] The above is an introduction to the method embodiment. The following is a further explanation of the scheme described in this application through an apparatus embodiment.

[0140] Figure 8 FIG. 8 is a block diagram of an iris image correction device 800 according to an embodiment of the present application. Figure 8 As shown, the device 800 includes:

[0141] An iris data acquisition module 810 is used to acquire iris data, wherein the iris data includes multiple groups of iris images, and each group of iris images has unique identification information;

[0142] An iris data preprocessing module 820, configured to preprocess the iris data and divide the preprocessed iris data into a training data set and a verification data set;

[0143] A deep neural network model construction module 830 is used to construct an iris image correction deep neural network model, and train the iris image correction deep neural network model according to the training data set, and verify it through the verification data set to obtain a trained iris image correction deep neural network model;

[0144] An iris image correction module 840 is used to input the iris image to be corrected into the trained iris image correction deep neural network model for forward reasoning to obtain a corrected iris image;

[0145] The linear normalization processing module 850 is used to perform linear normalization processing on the corrected iris image to obtain a corrected normalized image.

[0146] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0147] In the technical solution of this application, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0148] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0149] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. An iris image correction method, applied to iris images of any shape, characterized in that: include: Acquire iris data, wherein the iris data includes multiple groups of iris images, and each group of iris images has unique identification information; Preprocessing the iris data, and dividing the preprocessed iris data into a training data set and a verification data set; Constructing an iris image correction deep neural network model, and training the iris image correction deep neural network model according to the training data set, and verifying it through the verification data set to obtain a trained iris image correction deep neural network model; Inputting the iris image to be corrected into the trained iris image correction deep neural network model for forward reasoning to obtain a corrected iris image; Performing linear normalization processing on the corrected iris image to obtain a corrected normalized image; The preprocessing of the iris data comprises: Selecting an input image and a target image from the iris data; Preprocessing the input image to obtain a preprocessed input image, an input image mask, and a target mask of an output image; Preprocessing the target image to obtain a preprocessed target image and a target image mask; Preprocessing the input image to obtain a preprocessed input image, an input image mask, and a target mask of an output image, including: Positioning and segmenting the input image, identifying a valid iris area and an invalid iris area, and marking the invalid iris area as 0; Determine the center of the effective iris area, and cut the effective iris area based on the center of the effective iris area to obtain a standard cut area that can include all effective iris information; Adjusting the size of the standard cutting area and performing data standardization processing on all valid iris information in the standard cutting area to obtain a preprocessed input image; Obtaining the size information of an input image and creating a blank image having the same size information as the input image; Marking the valid iris region and the invalid iris region of the input image in the blank image to obtain an input image mask; The input image mask is subjected to concentric ring processing to obtain a target mask of the output image.

2. The method according to claim 1, characterized in that Selecting a target image from the iris data includes: Image analysis technology was used to evaluate the concentricity and clarity of the iris images in each group; The clear image in each group of iris images whose concentricity deviation from the ideal concentric circle state is the smallest and whose clarity reaches the set threshold is selected as the target image.

3. The method according to claim 1, characterized in that Preprocessing the target image to obtain a preprocessed target image and a target image mask includes: Positioning and segmenting the target image, identifying a valid iris area and an invalid iris area, and marking the invalid iris area as 0; Determine the center of the effective iris area, and crop the effective iris area based on the center of the effective iris area to obtain a standard cropped image, wherein the standard cropped image includes all effective iris information; Performing concentric circularization processing on the standard cropped image to obtain a concentric circularized image, wherein the concentric circularized image has the same geometric shape as the standard cropped image; The size of the concentrically annularized image is adjusted and all valid iris information in the concentrically annularized image is subjected to data standardization processing to obtain a preprocessed target image; Obtaining size information of a target image, and creating a blank image having the same size information as the target image; The valid iris region and the invalid iris region of the target image are marked in the blank image to obtain a target image mask.

4. The method according to claim 1, characterized in that: The method of constructing a deep neural network model for iris image correction includes: The U-shaped network structure is used to design the deep neural network model structure for iris image correction, wherein the U-shaped network structure includes an input processing module, an output processing module, a downsampling module, an upsampling module, and a jumper module, and the overall structure is a U-shaped symmetrical structure; Determining the training loss of the iris image correction deep neural network model structure by combining multiple losses with different weights; An iris image correction deep neural network model is constructed according to the iris image correction deep neural network model structure and the iris image correction deep neural network model structure training loss.

5. The method according to claim 4, characterized in that The method of determining the training loss of the iris image correction deep neural network model structure by combining multiple losses with different weights includes: Determine a joint loss function according to the loss values ​​calculated by the smoothL1 function, the L1loss function and the SSIM function and their corresponding loss calculation weights, and use the joint loss function as the training loss of the deep neural network model structure for iris image correction; Among them, the smoothL1 function is used to calculate the difference loss value between the output image and the target image; the L1loss function is used to calculate whether the mask of the actual output image is consistent with the mask target of the output image; the SSIM function is used to calculate the structural difference loss value between the output image and the target image.

6. The method according to claim 5, characterized in that The joint loss function is: ； in, represents the total joint loss value, Represents each loss item, For the The loss calculation weight, For the The calculated value of the loss.

7. The method according to claim 1, characterized in that The step of training the iris image correction deep neural network model according to the training data set includes: Step S91: input the training data set into the iris image correction deep neural network model for forward reasoning to obtain the current reasoning result; Step S92: Calculate the loss between the current inference result and the target mask of the output image through a joint loss function to obtain a current joint loss; Step S93: The current joint loss is transmitted back to the iris image segmentation deep neural network model through the reverse gradient propagation mechanism, and the current network weight is adjusted in combination with the learning rate; The steps S91 to S93 are executed cyclically until the training reaches a specified preset number of cycles, and the training ends.

8. An iris image correction device, applied to iris images of any shape, characterized in that: include: An iris data acquisition module, used to acquire iris data, wherein the iris data includes multiple groups of iris images, and each group of iris images has unique identification information; An iris data preprocessing module is used to preprocess the iris data and divide the preprocessed iris data into a training data set and a verification data set; the preprocessing of the iris data includes: Selecting an input image and a target image from the iris data; preprocessing the input image to obtain a preprocessed input image and an input image mask, and a target mask of an output image; Preprocessing the target image to obtain a preprocessed target image and a target image mask; Preprocessing the input image to obtain a preprocessed input image, an input image mask, and a target mask of an output image, including: Positioning and segmenting the input image, identifying the valid iris area and the invalid iris area, and marking the invalid iris area as 0; determining the center of the valid iris area, and cutting the valid iris area based on the center of the valid iris area to obtain a standard cutting area that can include all valid iris information; adjusting the size of the standard cutting area and performing data standardization processing on all valid iris information in the standard cutting area to obtain a preprocessed input image; obtaining the size information of the input image, and creating a blank image with the same size information; marking the valid iris area and the invalid iris area of ​​the input image in the blank image to obtain an input image mask; performing concentric ring processing on the input image mask to obtain a target mask of the output image A deep neural network model construction module, used to construct an iris image correction deep neural network model, and train the iris image correction deep neural network model according to the training data set, and verify it through the verification data set to obtain a trained iris image correction deep neural network model; An iris image correction module, used for inputting the iris image to be corrected into the trained iris image correction deep neural network model for forward reasoning to obtain a corrected iris image; The linear normalization processing module is used to perform linear normalization processing on the corrected iris image to obtain a corrected normalized image.

Images