Iris Texture Characterization Method Based on Multi-Directional Centrosymmetric Local Binary Pattern

Abstract

The present invention discloses an iris texture characterization method based on multi-directional center-symmetric local binary pattern. Aiming at the defect that it is difficult for the existing manually-designed descriptors to achieve balanced characterization of iris texture points and surfaces, the present invention combines the overall local neighborhood, the center, and the symmetric expression of the neighborhood, and designs an MDCS-LBP operator to characterize iris texture features. First, the weighted gray value of the overall local neighborhood is calculated, and then the coding rule is redefined to achieve coding between the weighted gray values to obtain the iris features of this locality. Finally, the iris features are dimension-reduced through binarization to improve the later recognition speed. Through experimental verification, this method can quickly and effectively characterize iris texture features, improve iris recognition performance, and obtain stable recognition effects for iris databases collected by different devices.

Description

Technical Field

[0001] The present invention belongs to the technical field of biometric recognition and information security, and particularly relates to an iris texture characterization method based on multi-directional center symmetric local binary pattern (MDCS-LBP). Background Art

[0002] Due to the advantages of universality, uniqueness, stability of the iris and the high complexity of iris texture, iris recognition technology is considered to be the most promising biometric recognition technology. In recent years, with the update of acquisition devices, iris recognition technology has developed rapidly. However, with the continuous update of requirements, iris recognition technology also needs to be continuously improved.

[0003] Iris recognition is a biometric recognition technology that uses computer vision to locate the iris in an image or video and identify its identity. Iris recognition mainly includes the following four steps: iris image acquisition and quality evaluation, iris image preprocessing, iris texture feature characterization and coding, and iris feature matching. Among them, iris texture characterization and coding is the core step of iris recognition, which directly determines the performance of iris recognition.

[0004] The core algorithm of the iris recognition system lies in the representation of iris texture. Currently, regarding the related research on iris texture representation, domestic and foreign scholars have proposed some related algorithms. Daugman et al. used 2D-Gabor filters to filter iris images to extract the phase features of iris texture, and encoded them according to the quadrants where the phase information is located, thus realizing the representation of iris texture (Daugman J G. Biometric personal identification system based on iris analysis: United States Patent, No. 5,291,560[P]. 1994.); the recognition performance of this method overly depends on the quality of iris images, and for iris images obtained by different acquisition devices, the filter parameters need to be adjusted to obtain better recognition performance. Li et al. manually designed the SCCS-LBP operator to represent iris texture features (Li Huanli, Guo Lihong, Li Xiaoming, Wang Xinzui, Dong Yuefang. Iris recognition based on statistically characteristic center symmetric local binary pattern[J]. Optics and Precision Engineering, 2013, 21(08): 2129-2136); this method effectively utilizes the distribution characteristics of iris texture and has achieved good results in both recognition speed and recognition accuracy, but the design of the SCCS-LBP operator ignores the role of the central pixel point and the relationship between multiple surrounding pixel points, and the anti-noise ability is not strong. Zhu et al. manually designed the MD-LBP operator to represent iris texture (Zhu Xiaodong, Zhang Qixian, Liu Yuanning, Wu Di, Wu Zukang, Wang Chaoqun, Li Xinlong. Iris recognition based on multi-directional local binary pattern and stable features[J]. Journal of Jilin University (Engineering and Technology Edition), 2021, 51(02): 650-658); this method considers the relationship between neighborhood pixel points and surrounding pixel points and provides more information for constructing feature mapping, but this method needs to obtain a stable feature recognition area through multiple filtering algorithms to improve the recognition effect, and there are significant differences in recognition performance for iris sample libraries with different resolutions. Summary of the Invention

[0005] An effective and robust representation of iris texture should consider both the overall local neighborhood, covering all directions, and the symmetric expression of the center and neighborhood to achieve a balanced representation of points and surfaces of the texture. Therefore, the present invention proposes a method for representing iris texture based on multi-directional center symmetric local binary pattern to improve the overall performance of iris recognition.

[0006] The method for representing iris texture based on multi-directional center symmetric local binary pattern is as follows:

[0007] Step (1). Select a 5*5 feature extraction template, calculate the weighted gray values in eight different directions and the central weighted gray value according to special calculation rules, and the calculated directional weighted gray values and the central weighted gray value constitute a feature coding template.

[0008] Step (2). Introduce the idea of central symmetric coding. To take into account the role of the central weighted gray value in texture characterization during coding, redefine the comparison rule between weighted gray values. The MDCS-LBP operator is composed of the calculation of weighted gray values and the redefined coding rule to realize the characterization of iris texture and obtain the MDCS-LBP feature map.

[0009] Step (3). Use the method of threshold binarization to reduce the dimension of the MDCS-LBP feature map.

[0010] Step (1). The specific method is as follows:

[0011] In the 5*5 feature extraction template, the pixel points in the selected direction and the central pixel point are weighted, and then the weighted gray values in each direction and the central weighted gray value are obtained by accumulating and averaging with their surrounding pixel points. The calculation formulas for the eight directional weighted gray values and the central weighted gray value are as follows:

[0012] 0° direction:

[0013] direction:

[0014] direction:

[0015] direction:

[0016] π direction:

[0017] direction:

[0018] direction:

[0019] direction:

[0020] center:

[0021] In the formula, P c is the central pixel value of the 5*5 feature extraction template, and P i (1≤i≤8) is based on P cThe pixel values on the inner ring of the 5*5 feature extraction template centered on Q j (1 ≤ j ≤ 16) are based on P c The pixel values on the outer ring of the 5*5 feature extraction template centered on G w (1 ≤ w ≤ 8) are based on P c The weighted gray values in each direction calculated centered on G c is the calculated central weighted gray value. The weighted gray values in each direction and the central weighted gray value calculated constitute the feature coding template.

[0022] Step (2). The specific method is as follows:

[0023] First, redefine the coding rule. Centered on the central weighted gray value G c , the feature coding template is divided into four groups of symmetric directions. For each group of symmetric directions, first compare the weighted gray values of two symmetric directions with respect to the central weighted gray value G c to generate the first 1-bit code, and then determine whether the central weighted gray value G c is between the weighted gray values of the two directions to generate the second 1-bit code. Each group of symmetric directions is represented by 2 bits of code. Starting from the horizontal direction for coding, the four groups of symmetric directions generate 8 bits of code, thereby characterizing the local iris texture information. The specific coding rule is as follows:

[0024]

[0025]

[0026]

[0027] The calculation rule of the MDCS-LBP operator is shown in Formula 10, where G w is the weighted gray value in the direction, G c is the central weighted gray value. In Formula 10, sgn(G w -G w+4 ) and α w are calculated as shown in Formula 11 and Formula 12. Texture characterization of the iris image is performed through the MDCS-LBP operator to obtain the MDCS-LBP feature map.

[0028] Step (3). The specific method is as follows:

[0029] When the MDCS-LBP operator characterizes iris texture information, 8-bit encoding is used to describe the changes in local iris texture information. The excessively high feature dimension will reduce the calculation efficiency and is also not conducive to the robustness of iris recognition. Therefore, threshold binarization is used to reduce the dimension of the MDCS-LBP feature map to obtain a binarized MDCS-LBP feature map. The specific dimension reduction calculation formula is as follows:

[0030]

[0031] Among them, MDCS-LBP(i,j) represents the pixel value at the (i,j) coordinate in the MDCS-LBP feature map.

[0032] The beneficial effects of the present invention are as follows:

[0033] Aiming at the defect that it is difficult for the existing manually designed descriptors to achieve balanced characterization of iris texture points and surfaces, the present invention combines the overall local neighborhood, the center, and the symmetric expression of the neighborhood to design an MDCS-LBP operator to characterize iris texture features. First, the weighted gray value of the overall local neighborhood is calculated, and then the encoding rule is redefined to achieve the encoding between the weighted gray values to obtain the iris features of this part. Finally, the iris features are dimension-reduced by binarization to improve the later recognition speed. Through experimental verification, this method can quickly and effectively characterize iris texture features, improve iris recognition performance, and obtain stable recognition effects for iris databases collected by different devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is the flowchart of the method of the present invention;

[0035] Figure 2 is the schematic diagram for calculating the weighted gray values in eight different directions and the center;

[0036] Figure 3 is the schematic diagram for encoding the weighted gray values; DETAILED DESCRIPTION OF THE INVENTION

[0037] The present invention will be further described below with reference to the accompanying drawings.

[0038] As Figure 1 shown, the specific steps of the method of the present invention are as follows:

[0039] The iris texture characterization method based on multi-directional center-symmetric local binary pattern is as follows:

[0040] Step (1). In order to provide more information for constructing the feature mapping, a 5*5 feature extraction template is selected, and the weighted gray values in eight different directions and the center weighted gray value are calculated according to special calculation rules. The calculated directional weighted gray values and the center weighted gray value constitute a feature encoding template.

[0041] As Figure 2 shown, in the 5*5 feature extraction template, the pixel points in the selected direction and the central pixel point are weighted, and then accumulated and averaged with the surrounding pixel points to obtain the weighted gray values in each direction and the central weighted gray value. The calculation formulas for the weighted gray values in eight directions and the central weighted gray value are as follows:

[0042] 0° direction:

[0043] Direction:

[0044] Direction:

[0045] Direction:

[0046] π direction:

[0047] Direction:

[0048] Direction:

[0049] Direction:

[0050] Center:

[0051] In the formula, P c is the central pixel value of the 5*5 feature extraction template, P i (1≤i≤8) is the pixel value on the inner ring of the 5*5 feature extraction template with P c as the center, Q j (1≤j≤16) is the pixel value on the outer ring of the 5*5 feature extraction template with P c as the center, G w (1≤w≤8) is the weighted gray value in each direction calculated with P c as the center, G c is the calculated central weighted gray value. The weighted gray values in each direction and the central weighted gray value calculated form the feature coding template.

[0052] Step (2). Introduce the idea of central symmetric coding. To take into account the role of the central weighted gray value in texture characterization during coding, redefine the comparison rule between weighted gray values. The MDCS-LBP operator is composed of the calculation of weighted gray values and the redefined coding rule to realize the characterization of iris texture and obtain the MDCS-LBP feature map.

[0053] The coding template formed by the calculated weighted gray values is as Figure 3 shown. With the central weighted gray value G c as the center, the feature coding template is divided into four groups of symmetric directions. For each group of symmetric directions, first compare the weighted gray values of two symmetric directions about the central weighted gray value G c to generate the first 1-bit coding, and then determine whether the central weighted gray value G c is between the weighted gray values of the two directions to generate the second 1-bit coding. Each group of symmetric directions is represented by 2-bit coding. Starting from G1 at the horizontal position, code in the direction indicated by the arrow. The four groups of symmetric directions generate 8-bit coding, thereby characterizing the local iris texture information. The specific coding rules are as follows:

[0054]

[0055]

[0056]

[0057] The calculation rule of the MDCS-LBP operator is shown in Equation 10, where G w is the directional weighted gray value, G c is the central weighted gray value. The calculations of sgn(G w -G w+4 ) and α w are shown in Equation 11 and Equation 12. Through the MDCS-LBP operator, texture characterization of the iris image is performed to obtain the MDCS-LBP feature map.

[0058] Step (3). Use the method of threshold binarization to reduce the dimension of the MDCS-LBP feature map.

[0059] When the MDCS-LBP operator characterizes iris texture information, 8-bit coding is used to describe the change of local iris texture information. Excessive feature dimensions will reduce the calculation efficiency and are also not conducive to the robustness of iris recognition. Therefore, threshold binarization is used to reduce the dimension of the MDCS-LBP feature map to obtain the binarized MDCS-LBP feature map. The specific dimension reduction calculation formula is as follows:

[0060]

[0061] Among them, MDCS-LBP(i, j) represents the pixel value at the coordinate (i, j) in the MDCS-LBP feature map.

[0062] It should be noted that the above-mentioned embodiments can be freely combined as needed. The above is only a detailed description of the preferred embodiments and principles of the present invention. For those of ordinary skill in the art, according to the idea provided by the present invention, there will be changes in the specific implementation manners, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. Iris texture characterization method based on multi-directional centrosymmetric local binary pattern, characterized in that The steps are as follows: Step (1): Select a 5*5 feature extraction template, calculate the weighted gray values in eight different directions and the central weighted gray value according to special calculation rules. The calculated directional weighted gray values and the central weighted gray value form a feature coding template; Step (2): Introduce the idea of central symmetric coding. To take into account the role of the central weighted gray value in texture representation during coding, redefine the comparison rules between weighted gray values; The MDCS-LBP operator is composed of the calculation of weighted gray values and the redefined coding rules to realize the representation of iris texture and obtain the MDCS-LBP feature map; Step (3): Use the method of threshold binaryzation to reduce the dimension of the MDCS-LBP feature map; Step (1): The specific method is as follows: In the 5*5 feature extraction template, the pixel points in the selected direction and the central pixel point are weighted, and then the weighted gray values in each direction and the central weighted gray value are obtained by accumulating and averaging with their surrounding pixel points. The calculation formulas for the eight directional weighted gray values and the central weighted gray value are as follows: Wherein, P c is the central pixel value of a 5*5 feature extraction template, P i is the pixel value on the inner ring of the 5*5 feature extraction template centered on P c , where 1≤i≤8; Q j is the pixel value on the outer ring of the 5*5 feature extraction template centered on P c , where 1≤j≤16; G w is the weighted gray value in each direction calculated with P c as the center, where 1≤w≤8; G c is the calculated central weighted gray value; the weighted gray values in each direction and the central weighted gray value calculated form a feature coding template.

2. The iris texture characterization method based on multi-directional centrosymmetric local binary pattern according to claim 1, characterized in that Step (2) The specific method is as follows: First, redefine the encoding rule with the central weighted gray value G c as the center, the feature encoding template is divided into four groups of symmetric directions. For each group of symmetric directions, first compare the weighted gray values of two symmetric directions with respect to the central weighted gray value G c to generate the first 1-bit code, and then determine whether the central weighted gray value G c is between the weighted gray values of the two directions to generate the second 1-bit code. Each group of symmetric directions is represented by 2 bits of code; starting from the horizontal direction for encoding, the four groups of symmetric directions generate 8 bits of code, thereby characterizing the local iris texture information. The specific encoding rule is as follows: The calculation rule of the MDCS-LBP operator is shown in Formula 10, where G w is the direction-weighted gray value, G c is the center-weighted gray value, sgn(G w -G w+4 ) and α w are calculated as shown in Formula 11 and Formula 12; Use the MDCS-LBP operator to represent the texture of the iris image to obtain the MDCS-LBP feature map.

3. The iris texture characterization method based on multi-directional centrosymmetric local binary pattern according to claim 2, characterized in that Step (3) The specific method is as follows: When the MDCS-LBP operator represents the iris texture information, 8-bit coding is used to describe the change of local iris texture information. Too high a feature dimension will reduce the calculation efficiency and is not conducive to the robustness of iris recognition. Therefore, threshold binaryzation is used to reduce the dimension of the MDCS-LBP feature map to obtain the binary MDCS-LBP feature map. The specific dimension reduction calculation formula is as follows: Where, MDCS-LBP(i,j) represents the pixel value at the (i,j) coordinate in the MDCS-LBP feature map.

Images