Rapid and accurate iris segmentation method based on target detection

Abstract

The invention relates to a target detection-based iris rapid and accurate segmentation method, which comprises the following steps of: obtaining an eye image for training, establishing a three-dimensional coordinate system and obtaining parameters of an iris ellipse by utilizing a three-dimensional mathematical relationship; constructing a directional bounding box fitted with the contour of the iris ellipse, the four edges of the directional bounding box being tangent to the iris ellipse, and solving the coordinates of the four vertexes of the directional bounding box; carrying out model training by taking the coordinates of the four vertexes of the directional bounding box and the parameters of the iris ellipse as model training labels so as to obtain a network model; the network model is used for detecting a directional bounding box surrounding an iris ellipse of the target eye image, and calculating and segmenting the iris ellipse according to the directional bounding box surrounding the iris. According to the method, a traditional iris segmentation mode depending on edge detection or pixel classification is abandoned, a brand new technical route based on deep learning target detection is adopted, and the limitation of a traditional method in processing a low-quality iris image is effectively overcome.

Description

Technical Field

[0001] The present invention relates to the field of biometric technologies, and more specifically, to a method for rapid and accurate segmentation of the iris based on object detection. Background Art

[0002] Iris recognition has the characteristic of high security in the field of biometrics, and its accurate segmentation of the iris is crucial. Early iris segmentation methods mainly relied on image processing techniques such as traditional edge detection or pixel classification techniques. Although these traditional methods performed well when processing high-quality images under controlled conditions, traditional methods faced many limitations when processing low-quality iris images, such as blurring, uneven illumination, occlusion, etc., resulting in poor segmentation effects and affecting the speed and accuracy of iris recognition. Therefore, it is necessary to develop new segmentation methods. Therefore, how to quickly and accurately segment the iris is exactly the technical problem to be solved by this application. Summary of the Invention

[0003] In view of the deficiencies of the prior art, the present invention proposes a technical solution for a method for rapid and accurate segmentation of the iris based on object detection, which can quickly and accurately locate the iris center and accurately extract the parameters of the iris ellipse, including the major axis, minor axis, and rotation angle, so as to improve the efficiency and accuracy of iris recognition and provide reliable data support for subsequent biometric identification and security authentication.

[0004] A method for rapid and accurate segmentation of the iris based on object detection includes the following processes: St10. Obtain the eye images for training, establish a three-dimensional coordinate system, and obtain the parameters of the iris ellipse using three-dimensional mathematical relationships; St20. Construct an oriented bounding box that fits the contour of the iris ellipse, with the four sides of the oriented bounding box tangent to the iris ellipse respectively, and solve the coordinates of the four vertices of the oriented bounding box; St30. Use the coordinates of the four vertices of the oriented bounding box and the parameters of the iris ellipse as the model training labels to train the model, thereby obtaining a network model; St40. The network model is used to detect the oriented bounding box that encloses the iris ellipse of the target eye image, and calculate and segment the iris ellipse based on the oriented bounding box that encloses the iris.

[0005] In summary, the above technical solution has the following beneficial effects: In this application, the four vertex coordinates of the iris elliptical oriented bounding box are first used as the model training labels, and the model is trained by combining the eye images and their corresponding labels. Then, the model can predict the oriented bounding box of the iris ellipse. Next, the iris ellipse parameters are calculated based on the four vertices of the oriented bounding box. Finally, the iris is accurately segmented according to the iris ellipse parameters. The beneficial effects of this application include: 1. Abandon the traditional iris segmentation method that relies on edge detection or pixel classification, and adopt a new technical route based on deep learning object detection. With the object detection algorithm (such as the YOLO algorithm) as the core, combined with custom parameter training and mathematical relationship verification functions, it effectively overcomes the limitations of traditional methods in dealing with low-quality iris images and realizes more accurate and rapid iris segmentation. 2. It can not only quickly and accurately locate the iris center, but also precisely extract all key parameters of the iris ellipse, including the major axis, minor axis, and rotation angle. While improving the iris recognition speed, it greatly improves the accuracy, providing more comprehensive and reliable data support for biometric identification and security authentication, and achieving a major breakthrough in data integrity and recognition performance compared with traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a schematic diagram of the steps of a method for rapid and accurate iris segmentation based on object detection; Figure 2 It is a schematic diagram of the flow chart of a method for rapid and accurate iris segmentation based on object detection; Figure 3 It is a schematic diagram of label calculation of a method for rapid and accurate iris segmentation based on object detection; Figure 4 It is a schematic diagram of the network model of a method for rapid and accurate iris segmentation based on object detection. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0007] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0008] As Figures 1 - 3 shown, a method for rapid and accurate iris segmentation based on object detection includes the following processes: St10. Obtain the eye images for training, establish a three-dimensional coordinate system, and use three-dimensional mathematical relationships to obtain the parameters of the iris ellipse.

[0009] St20. Construct an oriented bounding box that fits the elliptical contour of the iris. The four sides of the oriented bounding box are tangent to the iris ellipse respectively, and solve the coordinates of the four vertices of the oriented bounding box.

[0010] St30. Use the coordinates of the four vertices of the oriented bounding box and the parameters of the iris ellipse as model training labels to train the model, so as to obtain a network model.

[0011] St40. The network model is used to detect the oriented bounding box that encloses the iris ellipse in the target eye image, and calculate and segment the iris ellipse according to the oriented bounding box that encloses the iris.

[0012] In this application, first, the coordinates of the four vertices of the oriented bounding box of the iris ellipse are used as model training labels, and the model is trained by combining the eye image and its corresponding labels. Then, the model can predict the oriented bounding box of the iris ellipse. Next, the parameters of the iris ellipse are calculated based on the four vertices of the oriented bounding box. Finally, the iris is accurately segmented according to the parameters of the iris ellipse. The beneficial effects of this application include: 1. Abandon the traditional iris segmentation method that relies on edge detection or pixel classification, and adopt a new technical route based on deep learning object detection. With the object detection algorithm (such as the YOLO algorithm) as the core, combined with custom parameter training and mathematical relationship verification functions, it effectively overcomes the limitations of traditional methods in processing low-quality iris images and realizes more accurate and fast iris segmentation. 2. It can not only quickly and accurately locate the iris center, but also accurately extract all key parameters of the iris ellipse, including the major axis, minor axis and rotation angle. While improving the iris recognition speed, it greatly improves the accuracy, providing more comprehensive and reliable data support for biometric identification and security authentication, and achieving a major breakthrough in data integrity and recognition performance compared with traditional methods.

[0013] Specifically, both the eye images for training and the target eye images for recognition are preprocessed, including but not limited to steps such as denoising and enhancing contrast, to improve the image quality and facilitate subsequent object detection.

[0014] The method for obtaining the parameters of the iris ellipse is prior art. For example, a method for iris segmentation restricted by three-dimensional eyeball model parameters disclosed in the patent publication number CN118629083A discloses content such as establishing a three-dimensional coordinate through an eye image to obtain the eyeball center, iris radius parameters, iris ellipse function, and three-dimensional eyeball model. Based on the iris segmentation algorithm, the iris area in the eye image is detected, the iris center and iris area are obtained, and the iris center coordinates at the position where the difference between the minor axis and the major axis of the iris ellipse in the eye movement video is the smallest or relatively small are calculated as the eyeball center coordinates, and the major axis of the ellipse is taken as the iris radius.

[0015] Specifically, the process of obtaining the parameters of the iris ellipse includes the following: St11. Establish a three-dimensional coordinate system based on the eye image, with the positive direction of the X-axis to the right, the positive direction of the Y-axis upward, and the positive direction of the Z-axis outward, and obtain the center of the eyeball, iris radius parameter, iris ellipse function, three-dimensional eyeball model, etc. The eye movement can be regarded as rotating α degrees around the X-axis and then rotating β degrees around the Y-axis.

[0016] St12. Obtain the iris ellipse parameters by using the iris segmentation method restricted by the three-dimensional eyeball model parameters. The parameters of the iris ellipse include x, y, a, b, β, where x and y respectively represent the X-axis coordinate and Y-axis coordinate of the projection of the iris ellipse on the XY plane, a represents the major axis length, b represents the minor axis length, and β represents the rotation angle around the Y-axis.

[0017] The coordinates of the four vertices of the oriented bounding box are obtained through the following process: S21. According to the iris ellipse parameters, use its circumscribed matrix as the oriented bounding box. The coordinates of the four vertices of the oriented bounding box are represented in the following form: The coordinates of P1 are

[0018] The coordinates of P2 are

[0019] The coordinates of P3 are

[0020] The coordinates of P4 are

[0021] P1, P2, P3, and P4 respectively represent the four vertices of the bounding box. The professional term is Oriented Bounding Boxes Object Detection. The oriented bounding box (OBB) is a geometric structure used to describe the spatial range and direction of an object in the fields of object detection and computer vision. The OBB can be represented by the coordinates of four vertices, or determined by the center coordinates (x, y), width (width), height (height), and rotation angle (θ).

[0022] Use object detection to achieve iris segmentation, enhance the restriction of anatomical prior knowledge. The object detection uses the YOLOv8n-OBB model, which includes a backbone network, a neck network, and a head network.

[0023] The YOLOv8 model has the advantages of high efficiency and accuracy in the field of object detection. Utilizing its powerful feature extraction and object localization capabilities, it can accurately identify key objects in iris images, such as the rectangular box surrounding the iris. The YOLOv8-OBB model is specifically optimized for rotated bounding box detection, which is particularly important for iris segmentation because due to the shape and position of the iris, the bounding box may not be a standard axis-aligned rectangle, but requires a more flexible bounding box to accurately capture. In iris segmentation, this means that the model can more accurately identify and locate the boundaries of the iris, even if the iris is not facing the camera directly in the image.

[0024] According to the particularity of the iris segmentation task, relevant parameters of the YOLOv8-OBB model are custom-set. By adjusting the parameters, the model can better adapt to the characteristics of iris images, improve the processing ability for iris images of different qualities, and ensure more accurate identification of iris-related objects. The YOLOv8n-OBB model network with the smallest model parameters and relatively fast training speed is selected for training.

[0025] As Figure 4 shown, specifically, the YOLOv8n-OBB model includes a backbone network, a neck network, and a head network. The backbone network extracts features through a series of convolutional and deconvolutional layers, and uses residual connections and bottleneck structures to reduce the network size and improve performance. The neck network is responsible for multi-scale feature fusion and enhancement to improve the accuracy and robustness of object detection. The head network further processes the feature map to generate the specific information required for final object detection, including class, location, and confidence.

[0026] (1) Backbone network: The backbone network of YOLOv8n-OBB is responsible for feature extraction and consists of 5 convolutional (Conv) modules, 4 C2f models, and 1 spatial pyramid pooling fast (SPPF) module. It extracts features through a series of convolutional and deconvolutional layers, and uses residual connections and bottleneck structures to reduce the network size and improve performance.

[0027] The Conv module consists of a convolutional layer, a batch normalization layer (BN layer), and a SiLU activation function, which helps improve the learning efficiency and generalization ability of the network.

[0028] The C2f module has fewer parameters and better feature extraction capabilities. First, it uses a 1×1 convolutional kernel to reduce the number of input channels , to reduce the computational amount and memory consumption. Then, use the Bottleneck block for convolution operations to extract feature information. Next, use residual connections to directly add the input to the output to form a cross-layer connection. Finally, use a 1×1 convolutional kernel to restore the number of channels of the feature map.

[0029] The SPPF module performs pooling operations of multiple different sizes, splices feature maps of different scales together, and then obtains the final feature map through 1×1 convolution. This helps to enhance the model's detection ability for targets of different sizes.

[0030] In the Backbone part of the YOLOv8n-OBB model, the convolution module, C2f module, and SPPF module all have specific parameter settings, such as the number of filters, convolutional kernel size, stride, etc.

[0031] The specific model parameters are as follows: 0 Conv [3, 16, 3, 2] 1 Conv [16, 32, 3, 2] 2 C2f [32, 32, 1, True] 3 / P3 Conv [32, 64, 3, 2] 4 C2f [64, 64, 2, True] 5 / P4 Conv [64, 128, 3, 2] 6 C2f [128, 128, 2, True] 7 / P5 Conv [128, 256, 3, 2] 8 C2f [256, 256, 1, True] 9 SPPF [256, 256, 5] (2) Neck network: The Neck part of the YOLOv8n-OBB model is responsible for multi-scale feature fusion and enhancement to improve the accuracy and robustness of object detection. Specifically, the Neck part inputs the feature maps of different stages of the Backbone into the Path Aggregation Network (PAN-FPN) with a Feature Pyramid Network for feature fusion. Input P3, P4, and P5 of the Backbone into the PAN-FPN network structure to fuse feature maps of multiple scales; upsample P5 and fuse it with P4 to get F1, fuse F1 through a C2f layer and then upsample it with P3 to get T1, fuse T1 through a convolutional layer with F1 to get F2, fuse F2 through a C2f layer to get T2, fuse T2 through a convolutional layer with P5 to get F3, fuse F3 through a C2f layer to get T3. Finally, T1, T2, and T3 are the outputs of the neck network.

[0032] The specific model parameters are as follows: Number of layers Module name Parameters 10 Upsampling [None, 2, 'nearest'] 11 Concatenation [-1,6] 12 C2f [384, 128, 1] 13 Upsampling [None, 2, 'nearest'] 14 Concatenation [-1,4] 15 C2f [192, 64, 1] 16 Conv [64, 64, 3, 2] 17 Concatenation [-1,12] 18 C2f [192, 128, 1] 19 Conv [128, 128, 3, 2] 20 Concatenation [-1,9] 21 C2f [384, 256, 1] (3) Head Network (Head): The Head part of YOLOv8n-OBB further processes the feature map output by the Neck to generate the specific information required for final object detection, including class, location, and confidence. Specifically, YOLOv8-OBB uses a decoupled head approach to generate the prediction feature map. The decoupled head processes the classification task and the regression task using different network layers. Each branch includes 1 Conv module and 1 convolutional layer. The classification task uses an independent head to predict class probabilities, and the regression task uses another independent head to predict bounding boxes. This design allows the model to focus more on their respective tasks and improves the accuracy of predictions. The final prediction results will undergo non-maximum suppression to remove duplicate detection boxes. The bounding box with the highest confidence is retained, and other bounding boxes with high overlap are removed to ensure that each object is detected only once. Since YOLOv8-OBB is specifically designed for rotated object detection, its head network can predict rotated bounding boxes, which include the center point coordinates, width, height, and rotation angle.

[0033] The specific model parameters are as follows: Number of layers Module name Parameters (number of layers) 22 OBB [15, 18, 21] Use the coordinate parameters of the four vertices of the bounding box as the labels for training eye images, and train the YOLOv8n-OBB model.

[0034] St31. Follow the classification loss function of the YOLOv8n-OBB model , and the distribution loss function .

[0035] St32. In the bounding box loss function , combined with the anatomical knowledge of the eyeball, introduce the three-dimensional position distance loss function , which is used to calculate the cosine similarity between the optical axis endpoints of the prediction and the target parameters, so as to ensure the accuracy of segmentation. The formula is as follows:

[0036] where is used to calculate the probability intersection over union between oriented bounding boxes. The formula is as follows:

[0037] where obb1 and obb2 are the true bounding box and the predicted bounding box respectively, a1, b1, c1 are the covariance matrix elements calculated from obb1, and a2, b2, c2 are the covariance matrix elements calculated from obb2. is a positive number used to avoid division by zero;

[0038] Among them, and respectively represent the positions of the predicted iris ellipse and the real iris ellipse in three-dimensional space. The positions of the iris ellipses in three-dimensional space are obtained through the following formula: , , , ,

[0039] Where and are the coordinates of the center point of the iris ellipse, a represents the major axis of the iris ellipse, represents the position of the iris ellipse;; St33. The final three-dimensional loss function is , , The sum of

[0040] The three-dimensional loss function Loss is calculated through the above formula.

[0041] St41. The network model is used to detect the oriented bounding box of the target eye image, and the coordinates of its four vertices are respectively represented as , , , ; The parameters of the iris ellipse include the coordinates of the iris center point , the major axis length , the minor axis length , the rotation angle around the Y axis , and are represented in the following form: ; ; ; ; ; ; St42. Segment the iris region of the eye image according to the iris ellipse parameters.

[0042] In this application, the four vertex coordinates of the iris elliptical orientation bounding box are first used as the model training labels. The YOLOv8n-OBB model is trained by combining the eye images and their corresponding labels, and a three-dimensional position distance loss function is introduced to make the model detection results more conform to the anatomical prior knowledge. The model can not only quickly and accurately locate the iris center, but also precisely extract all the key parameters of the iris ellipse, including the major axis, minor axis and rotation angle, and can segment the iris.

[0043] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A fast and accurate iris segmentation method based on target detection, characterized in that: The process includes: St10, obtaining an eye image for training, establishing a three-dimensional coordinate system and obtaining the parameters of the iris ellipse using a three-dimensional mathematical relationship; St20, construct an oriented bounding box that fits the iris ellipse contour, where the four sides of the oriented bounding box are tangent to the iris ellipse respectively, and solve the coordinates of the four vertices of the oriented bounding box; St30, using the coordinates of the four vertices of the oriented bounding box and the parameters of the iris ellipse as model training labels for model training, thereby obtaining a network model; St40, the network model is used to detect the directional bounding box surrounding the iris ellipse of the target eye image, and calculate and segment the iris ellipse based on the directional bounding box surrounding the iris.

2. The method for fast and accurate iris segmentation based on target detection according to claim 1, characterized in that: The coordinates of the four vertices of the bounding box are obtained by the following process: St11, establish a three-dimensional coordinate system based on the eye image, with the X-axis pointing rightward as the positive direction, the Y-axis pointing upward as the positive direction, and the Z-axis pointing outward as the positive direction, to obtain the eyeball center, iris radius parameter, iris ellipse function and three-dimensional eyeball model; St12. The iris segmentation method based on the 3D eyeball model parameter constraints obtains the iris ellipse parameters. The iris ellipse parameters include x, y, a, b, β, where x and y represent the X-axis coordinate and Y-axis coordinate of the iris ellipse projected on the XY plane, a represents the length of the major axis, b represents the length of the minor axis, and β represents the rotation angle around the Y axis; S21. According to the iris ellipse parameters, its circumscribed matrix is ​​used as an oriented bounding box, and the coordinates of the four vertices of the oriented bounding box are expressed in the following form: The coordinates of P1 are , The coordinates of P2 are , The coordinates of P3 are , The coordinates of P4 are , P1, P2, P3, and P4 represent the four vertices of the bounding box respectively.

3. The method for fast and accurate iris segmentation based on target recognition according to claim 2, characterized in that: Target detection is used to achieve iris segmentation and enhance the constraints of anatomical prior knowledge. Target detection adopts the YOLOv8n-OBB model, which includes a backbone network, a neck network, and a head network.

4. The method for fast and accurate iris segmentation based on target detection according to claim 3, characterized in that: Introduce the three-dimensional distance loss function into the YOLOv8n-OBB model to form the final three-dimensional loss function; St31, using the classification loss function of the YOLOv8n-OBB model , distribution loss function ; St32, in the bounding box loss function In the paper, combined with the anatomical knowledge of the eyeball, the three-dimensional position distance loss function is introduced , which is used to calculate the cosine similarity between the optical axis endpoints between the predicted and target parameters to ensure the accuracy of the segmentation. The formula is as follows: ， in, It is used to calculate the probability intersection-over-union ratio between oriented bounding boxes, and its formula is as follows: ， Among them, obb1 and obb2 are the real bounding box and the predicted bounding box respectively, a1, b1, c1 are the covariance matrix elements calculated by obb1, and a2, b2, c2 are the covariance matrix elements calculated by obb2. is a positive number used to avoid division by zero; ， in, , They represent the positions of the predicted iris ellipse and the real iris ellipse in three-dimensional space respectively. The position of the iris ellipse in three-dimensional space is obtained by the following formula: ， ， ， ， ， in , is the coordinate of the center point of the iris ellipse, a represents the major axis of the iris ellipse, Indicates the position of the iris ellipse; St33, the final three-dimensional loss function is , , the sum of , The three-dimensional loss function Loss is calculated by the above formula.

5. The method for fast and accurate iris segmentation based on target detection according to claim 3, characterized in that: St41, the network model is used to detect the oriented bounding box of the target eye image, and its four vertex coordinates are respectively expressed as , , , ; The parameters of the iris ellipse include the coordinates of the iris center point , major axis length , short axis length , Rotation angle around the Y axis , and is expressed in the following form: ; ； ； ； ； ； St42, based on the iris ellipse parameters, the iris area of ​​the eye image is segmented.

Images