Training method and recognition method of iris recognition model for eyelashes

By segmenting and correcting the area ratio of the iris image and adjusting the iris training set of the input CNN neural network, the problem of low recognition accuracy of iris images with eyelashes is solved, and higher recognition accuracy and effectiveness are achieved.

CN114463831BActive Publication Date: 2024-11-05SHENZHEN MAXVISION TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210017354.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-08
Publication Date
2024-11-05
Estimated Expiration
2042-01-08

AI Technical Summary

Technical Problem

When existing iris recognition technology processes iris images with eyelashes, the recognition accuracy is low and cannot effectively solve the recognition problems caused by eyelash occlusion.

Method used

By acquiring the iris image, segmenting the iris and eyelash areas, calculating the area ratio, setting the threshold to correct the image, inputting the CNN neural network for training, using the one-hot function and softmax activation function for category tag encoding, adjusting the iris image to reduce eyelash occlusion Influence, an iris recognition model with eyelashes was trained.

Benefits of technology

It improves the recognition accuracy of the iris images with eyelashes, ensures the effectiveness of the training set, prevents excessive eyelashes from affecting the recognition process, and effectively solves the recognition problem of iris images with eyelashes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114463831B_ABST
    Figure CN114463831B_ABST
Patent Text Reader

Abstract

The present application discloses a training method for an iris recognition model for eyelashes, which comprises: obtaining an iris training set including a plurality of iris images; performing image segmentation on each iris image in the iris training set to obtain an iris region H and an eyelash region M; respectively calculating the iris pixel area S of the iris region H of each iris image; H and eyelash pixel area S of eyelash region M M , and calculate the eyelash pixel area S of the iris image M and the iris pixel area S H The area ratio is denoted as P, P = S M / S H ; Set a threshold value P0, compare the area ratio P of each iris image in the iris training set with the threshold value P0, put all iris images in the iris training set into the correction: P<P0; and input the corrected iris training set into the CNN neural network for learning and training of the iris recognition model. The present application also provides an iris recognition method for eyelashes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of image processing, and more specifically, to a training method and a recognition method for an iris recognition model for eyelashes. Background Art

[0002] Existing iris recognition technologies are mainly divided into two categories. The first category is based on traditional pattern recognition methods, that is, manually designing features to extract image features and connecting them to a classifier to obtain classification results. The second category is based on deep learning CNN classification methods, and the extraction process is data-driven.

[0003] Existing iris recognition technologies, whether early traditional pattern recognition or recent deep learning, are based on relatively clean iris data without eyelashes for better recognition results. Considering that in recent years, people, especially women, have trimmed their original eyelashes or even had false eyelashes for reasons such as beauty and dressing up, this poses a new challenge to iris recognition. The iris recognition model trained with iris data without eyelashes does not perform well for irises with eyelashes, resulting in a significant decrease in the accuracy of the iris recognition model in such cases. Summary of the invention

[0004] In view of the prior art, the technical problem solved by the present application is to provide a training method and a recognition method for an iris recognition model for eyelashes. The training method and the recognition method are conducive to training an iris recognition model for eyes with eyelashes, and are conducive to solving the iris recognition problem when eyelashes are present.

[0005] In order to solve the above technical problems, in a first aspect, the present application provides a training method for an iris recognition model for eyelashes, comprising:

[0006] Obtain an iris training set including a plurality of iris images;

[0007] Performing image segmentation on each iris image in the iris training set to obtain an iris region H and an eyelash region M;

[0008] Calculate the iris pixel area S of the iris region H of each iris image respectively H and eyelash pixel area S of eyelash region M M , and calculate the eyelash pixel area S of the iris image M and the iris pixel area S H The area ratio is denoted as P, P = S M / S H ;

[0009] Setting a threshold value P0, comparing the area ratio P of each iris image in the iris training set with the threshold value P0, and correcting all iris images corresponding to P≥P0 in the iris training set to P<P0; and,

[0010] The modified iris training set is input into the CNN neural network for learning and training of the iris recognition model.

[0011] In a possible implementation, the step of determining a threshold value P0, comparing the area ratio P of each iris image in the iris training set with the threshold value P0, and correcting all iris images corresponding to P≥P0 in the iris training set to P<P0 includes:

[0012] Selecting an iris image C with an area ratio P≥P0 and an iris image D with an area ratio P<P0 in the iris training set;

[0013] The iris image C is segmented to obtain the iris region and the eyelash region, which are denoted as H C and M C , the iris image D is segmented to obtain the iris region and eyelash region, which are denoted as H D and M D ;

[0014] Transplant the eyelashes in the eyelash area of ​​iris image C to the eyelash area of ​​iris image D, and generate a new iris image F composed of the iris in iris image D and the eyelashes in iris image C;

[0015] The iris image C in the iris training set is replaced by the new iris image F.

[0016] In a possible implementation, when the modified iris training set is input into the CNN neural network for learning and training, the category label is encoded based on a one-hot function with a vector dimension of K:

[0017] The category label encoding of any iris image t in the iris training set after correction is set using the one-hot function:

[0018] Among them, P t is the area ratio of the iris pixel area to the eyelash pixel area of ​​the iris image t, P t The negative exponential function value of .

[0019] In a possible implementation, when the modified iris training set is input into the CNN neural network for learning and training the iris model, softmax is set as the training activation function, and the category confidence of the iris image t in the CNN neural network is set to

[0020] In one possible implementation,

[0021]

[0022] Among them, i represents the pixel point belonging to the iris area, and j represents the pixel point belonging to the eyelash area.

[0023] In a possible implementation, P0 is set to 10%.

[0024] In a possible implementation, the PSPNet segmentation model is used to perform image segmentation on each iris image in the iris training set.

[0025] In a possible implementation, the CNN neural network is a network based on a mobilenet model structure.

[0026] In the training method of the iris recognition model for eyelashes, by setting a threshold value P0, each iris image in the iris training set used for CNN neural network training of the iris recognition model is adjusted so that the area ratio of the eyelash region to the iris region is less than P0, so as to ensure the validity of the iris images in the iris training set, thereby training an iris recognition model for solving the problem of eyelashes, which is beneficial to solving the problem of recognizing iris images with eyelashes.

[0027] In a second aspect, the present application provides an iris recognition method for eyelashes, comprising:

[0028] Obtain the iris image of the person to be identified J, perform image segmentation on the iris image of the identified person J, and obtain the iris area H j and eyelash area M j ;

[0029] Set the threshold P1 and calculate the eyelash area M j and iris area H j The area ratio P j , if P j ≥P1, then reacquire the iris image of the person to be identified J until the iris image area of ​​the person to be identified J is larger than P j <P1;

[0030] Iris recognition is performed using an iris recognition model trained using the training method for an iris recognition model for eyelashes.

[0031] In a possible implementation, the P1 is set to 10%.

[0032] In the iris recognition method for eyelashes, by setting a threshold value P0, each iris image in the iris training set used for CNN neural network iris recognition model training is adjusted to have an area ratio of eyelash area to iris area less than P0, so as to ensure the effectiveness of the iris images in the iris training set, thereby training an iris recognition model for solving the problem of eyelashes, which is conducive to solving the problem of identifying iris images with eyelashes; and by setting a threshold value P1, it is possible to effectively prevent the eyelashes of the iris image of the person to be identified J from being too long, thereby affecting the iris recognition process. Therefore, the iris recognition method for eyelashes can effectively solve the problem of identifying iris images with eyelashes in reality. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0034] Figure 1 A flowchart of a method for training an iris recognition model for eyelashes according to an embodiment of the present application;

[0035] Figure 2 A flowchart of the steps of correcting all iris images corresponding to P≥P0 in the iris training set to P<P0 in an embodiment of the present application;

[0036] Figure 3 It is a schematic diagram of obtaining eyelash regions by performing image segmentation on an iris image C with an area ratio P≥P0 according to an embodiment of the present application;

[0037] Figure 4 An iris image D with an area ratio P<P0 selected in the embodiment of the present application;

[0038] Figure 5 For the embodiment of this application, Figure 3 Eyelash transplantation to Figure 4 Schematic diagram of the result of generating an iris image F in the eyelash area;

[0039] Figure 6 This is a flow chart of an iris recognition method for eyelashes according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0041] It should be noted that when an element is referred to as being "fixed to" or "disposed on" another element, it can be directly on the other element or indirectly on the other element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element.

[0042] It should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0043] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0044] The training method of the iris recognition model for eyelashes and the iris recognition method system for eyelashes of the present application are now described in detail with reference to the accompanying drawings.

[0045] Reference Figure 1 The training method of the iris recognition model for eyelashes provided in the embodiment of the present application comprises the following steps:

[0046] Step S10: obtaining an iris training set including a plurality of iris images;

[0047] Step S11: performing image segmentation on each iris image in the iris training set to obtain an iris region H and an eyelash region M;

[0048] Step S12: Calculate the iris pixel area S of the iris region H of each iris image H and eyelash pixel area S of eyelash region M M , and calculate the eyelash pixel area S of the iris image M and the iris pixel area S H The area ratio is denoted as P, P = S M / S H ;

[0049] Step S12: setting a threshold value P0, comparing the area ratio P of each iris image in the iris training set with the threshold value P0, and correcting all iris images corresponding to P≥P0 in the iris training set to P<P0;

[0050] Step S14: input the modified iris training set into the CNN neural network to perform learning and training of the iris recognition model.

[0051] In the training method of the iris recognition model for eyelashes, by setting a threshold value P0, each iris image in the iris training set used for CNN neural network training of the iris recognition model is adjusted so that the area ratio of the eyelash region to the iris region is less than P0, so as to ensure the validity of the iris images in the iris training set, thereby training an iris recognition model for solving the problem of eyelashes, which is beneficial to solving the problem of recognizing iris images with eyelashes.

[0052] It is understandable that when training an iris recognition model, if the iris images in the training set are taken when the eyelashes cover most of the iris area or completely cover the iris area, it will affect the extraction of iris features during the iris recognition process to a certain extent, thereby affecting the training of an effective iris recognition model.

[0053] In step S10, all iris images in the iris training set have the same size ratio and are all irises with eyelashes, and in each iris image, the iris and eyelashes basically occupy the entire image. The iris training set includes iris images with eyelashes and iris images without eyelashes. The iris images with eyelashes include iris images with an area ratio greater than or equal to P0 and iris images with an area ratio less than P0. The diversity of data can increase the adaptability of the trained iris recognition model.

[0054] In step S11, the PSPNet segmentation model is used to perform image segmentation on each iris image in the iris training set, and the PSPNet method is used to perform semantic segmentation on each iris image in the iris training set to obtain an iris region H and an eyelash region M.

[0055] In step S12, the iris pixel area S of the iris region H of each iris image is calculated. H and eyelash pixel area S of eyelash region M M The formula is as follows:

[0056]

[0057] Among them, i represents the pixel point belonging to the iris area, and j represents the pixel point belonging to the eyelash area.

[0058] Reference Figure 2In step S13, a threshold value P0 is determined, and the area ratio P of each iris image in the iris training set is compared with the threshold value P0, and the step of correcting all iris images corresponding to P≥P0 in the iris training set to P<P0 includes:

[0059] Step S131: selecting an iris image C with an area ratio P≥P0 and an iris image D with an area ratio P<P0 in the iris training set;

[0060] Step S132: Segment the iris image C to obtain an iris region and an eyelash region, which are respectively denoted as H C and M C , the iris image D is segmented to obtain the iris region and eyelash region, which are denoted as H D and M D ;

[0061] Step S133: transplanting the eyelashes in the eyelash area of ​​iris image C to the eyelash area of ​​iris image D, generating a new iris image F composed of the iris of iris image D and the eyelashes of iris image C;

[0062] Step S134: replacing the iris image C in the iris training set with the new iris image F.

[0063] Figure 3 is a schematic diagram of obtaining the eyelash area by segmenting an iris image C with an area ratio P≥P0, Figure 4 is the iris image D with area ratio P<P0, Figure 5 For the general Figure 3 Eyelash transplantation to Figure 4 Schematic diagram of the result of generating an iris image F in the eyelash area.

[0064] In one embodiment of the application, P0 is set to 10%. In other embodiments, P0 can be set to other values ​​and can be adjusted according to actual conditions.

[0065] In step S14, when the modified iris training set is input into the CNN neural network for learning and training, the category label is encoded based on a one-hot function with a vector dimension of K:

[0066] The category label encoding of any iris image t in the iris training set after correction is set using the one-hot function:

[0067] Among them, P t is the area ratio of the iris pixel area to the eyelash pixel area of ​​the iris image t, P t The negative exponential function value of .

[0068] Furthermore, in step S500, during the learning and training process of the iris recognition model, softmax is set as the training activation function, and the category confidence of the iris image t in the CNN neural network is set to In one embodiment of the application, the CNN neural network is a network based on the mobilenet model structure.

[0069] Understandably, because the longer the eyelashes are, the greater the impact on the accuracy of iris recognition, which means that the judgment result of each image in the training set is less reliable and the recognition result is less reliable during the training process. Thus, during the CNN neural network training process, a category confidence of the area ratio P of any iris image t in the iris training set is set, and the category confidence is used to continuously influence the training process, so as to train a more convincing iris recognition model, that is, to train a more accurate iris recognition model. Therefore, the above-set category confidence can be used to add the impact of eyelashes on the training process, so as to be more in line with the actual situation, so as to ensure that a more accurate iris recognition model is trained.

[0070] Reference Figure 6 The iris recognition method for eyelashes provided in the embodiment of the present application comprises the following steps:

[0071] Step S20: Obtain an iris image of the person to be identified J, perform image segmentation on the iris image of the identified person J, and obtain an iris region H j and eyelash area M j ;

[0072] Step S21: Set the threshold value P1 and calculate the eyelash area M j and iris area H j The area ratio P j , if P j ≥P1, then reacquire the iris image of the person to be identified J until the iris image area of ​​the person to be identified J is larger than P j <P1;

[0073] Step S22: performing iris recognition using the iris recognition model trained by the training method for the iris recognition model for eyelashes.

[0074] In one embodiment of the application, the P1 is set to 10%.

[0075] In this embodiment, by setting the threshold value P0, each iris image in the iris training set used for CNN neural network iris recognition model training is adjusted to have an area ratio of the eyelash area to the iris area less than P0, so as to ensure the effectiveness of the iris images in the iris training set, thereby training an iris recognition model for solving the problem of eyelashes, which is conducive to solving the problem of identifying iris images with eyelashes; and by setting the threshold value P1, it is possible to effectively prevent the eyelashes of the iris image of the person to be identified J from being too long, thereby affecting the iris recognition process. Therefore, the iris recognition method for eyelashes can effectively solve the problem of identifying iris images with eyelashes in reality.

[0076] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A training method for an iris recognition model for eyelashes, characterized in that: include: Obtain an iris training set including a plurality of iris images; Performing image segmentation on each iris image in the iris training set to obtain an iris region H and an eyelash region M; Calculate the iris pixel area S of the iris region H of each iris image respectively H and eyelash pixel area S of eyelash region M M , and calculate the eyelash pixel area S of the iris image M and the iris pixel area S H The area ratio is denoted as P, P = S M / S H ; Setting a threshold value P0, comparing the area ratio P of each iris image in the iris training set with the threshold value P0, and correcting all iris images corresponding to P≥P0 in the iris training set to P<P0; as well as Inputting the modified iris training set into the CNN neural network to perform learning and training of the iris recognition model; The threshold value P0 is set, the area ratio P of each iris image in the iris training set is compared with the threshold value P0, and all iris images corresponding to P≥P0 in the iris training set are corrected to P<P, including: Selecting an iris image C with an area ratio P≥P0 and an iris image D with an area ratio P<P0 in the iris training set; The iris image C is segmented to obtain the iris region and eyelash region, which are denoted as H C and M C , the iris image D is segmented to obtain the iris region and eyelash region, which are denoted as H D and M D ; Transplant the eyelashes in the eyelash area of ​​iris image C to the eyelash area of ​​iris image D, and generate a new iris image F composed of the iris in iris image D and the eyelashes in iris image C; The iris image C in the iris training set is replaced by the new iris image F.

2. The method for training an iris recognition model for eyelashes according to claim 1, characterized in that: When the modified iris training set is input into the CNN neural network for learning and training, the category label is encoded based on the one-hot function with a vector dimension of K: The category label encoding of any iris image t in the iris training set after correction is set using the one-hot function: Among them, P t is the area ratio of the iris pixel area to the eyelash pixel area of ​​the iris image t, P t The negative exponential function value of .

3. The method for training an iris recognition model for eyelashes according to claim 2, characterized in that: In the process of inputting the modified iris training set into the CNN neural network to learn and train the iris model, softmax is set as the training activation function, and the category confidence of the iris image t in the CNN neural network is set to 4. The method for training an iris recognition model for eyelashes according to claim 1, characterized in that: Among them, i represents the pixel point belonging to the iris area, and j represents the pixel point belonging to the eyelash area.

5. The method for training an iris recognition model for eyelashes according to claim 1, characterized in that: The P0 is set to 10%.

6. The method for training an iris recognition model for eyelashes according to claim 1, characterized in that: The PSPNet segmentation model is used to perform image segmentation on each iris image in the iris training set.

7. The method for training an iris recognition model for eyelashes according to claim 1, characterized in that: The CNN neural network is a network based on the mobilenet model structure.

8. An iris recognition method for eyelashes, characterized in that: Obtain the iris image of the person to be identified J, perform image segmentation on the iris image of the identified person J, and obtain the iris area H j and eyelash area M j ; Set the threshold P1 and calculate the eyelash area M j and iris area H j The area ratio P j , if P j ≥P1, then reacquire the iris image of the person to be identified J until the iris image area of ​​the person to be identified J is larger than P j <P1; and Iris recognition is performed using an iris recognition model trained by the training method for an iris recognition model for eyelashes as described in any one of claims 1 to 7.

9. The iris recognition method for eyelashes according to claim 8, characterized in that: The P1 is set to 10%.