A real-time retrieval method for large-scale finger vein images

Abstract

The present invention discloses a large-scale finger vein image real-time retrieval method, comprising: 1) obtaining finger vein images to be identified and a large-scale finger vein database and normalizing them; 2) performing image enhancement processing on the normalized finger vein images; 3) obtaining the finger vein image to be identified after enhancement processing in step 2), obtaining the region with the richest feature information, and performing horizontal and vertical projection on the region to obtain a regional grayscale projection vector; 4) obtaining the finger vein image of the finger vein database after enhancement processing in step 2), and obtaining a regional grayscale projection vector; 5) calculating the similarity between the regional projection vector of the image to be identified and multiple regional grayscale projection vectors of each finger vein image in the database, and taking the value with the highest similarity of each finger vein image for sorting; 6) extracting multi-scale HOG feature matching according to the sorting result, and returning the finger vein image number with the highest matching degree as the finger vein image real-time retrieval result. The present invention effectively reduces the retrieval time and achieves real-time performance.

Description

Technical Field

[0001] The present invention relates to the technical field of finger vein recognition and information security, and in particular to a large-scale finger vein image real-time retrieval method. Background Art

[0002] Finger vein recognition is a biometric recognition technology that collects the distribution image of the veins inside the finger for identity recognition. It has four characteristics: liveness detection, internal characteristics, non-contact, and simple collection equipment. Therefore, it is difficult to be forged and has a high level of recognition security. It is widely used in security systems, public places, community access control and other living and working places.

[0003] With the widespread application of finger veins, the scale of finger vein databases collected in some occasions or fields is getting larger and larger. For the identification and matching of finger vein images to be identified, it is necessary to retrieve the most similar results from a large-scale finger vein database, which means that thousands or tens of thousands of finger vein images in the database must be compared one by one. Due to the large scale of finger vein data, the time-consuming feature extraction of finger vein images, and the high feature dimension, traditional finger vein recognition retrieval technology cannot meet real-time requirements. Furthermore, considering that in some places or fields, the finger vein recognition task needs to be transplanted to embedded devices, and the computing power of embedded devices is low, it takes longer, which will further affect the user experience. Therefore, the realization of real-time retrieval in large-scale finger vein images is an urgent problem to be solved.

[0004] In summary, a large-scale finger vein image real-time retrieval method is invented, which has high practical application value. Summary of the invention

[0005] The purpose of the present invention is to overcome the problem in the prior art that large-scale finger vein image retrieval is time-consuming and cannot meet the real-time requirements, and proposes a large-scale finger vein image real-time retrieval method to reduce the retrieval time and achieve real-time performance.

[0006] To achieve the above object, the present invention provides a technical solution: a large-scale finger vein image real-time retrieval method, comprising the following steps:

[0007] 1) Obtain the finger vein image to be identified and the large-scale finger vein database, normalize the image, and obtain an image size of w×h, where the length is w and the width is h;

[0008] 2) performing image enhancement processing on the finger vein image normalized in step 1);

[0009] 3) The finger vein image to be identified after enhancement processing is obtained from step 2) and the size is The rectangle traverses the image to obtain the area with the richest feature information, and obtains the center point (a, b) of the area with the richest feature information. The area is horizontally and vertically projected to obtain the regional grayscale projection vector q;

[0010] 4) Take (a, b) in step 3) as the center to obtain the neighborhood range, and the point in the neighborhood range is (a0, b0), where Get the finger vein images of the finger vein database after the enhancement processing in step 2), and in each image, use the size of The rectangular area with the center (a0, b0) is moved and traversed and horizontally and vertically projected to obtain multiple regional grayscale projection vectors p for each image. i ;

[0011] 5) The p of each finger vein image in the finger vein database obtained in step 4) is i Calculate the similarity with q obtained in step 3), and sort the finger vein images by the value with the highest similarity;

[0012] 6) According to the sorting result of step 5), the corresponding number of the finger vein image whose ranking is less than the threshold threshold is taken as a candidate, and the image numbered as a candidate in the finger vein image of the finger vein database enhanced by step 2) is selected, and the multi-scale HOG features are extracted and matched with the finger vein image to be identified enhanced by step 2), and the finger vein image number with the highest matching degree is returned as the real-time retrieval result of the finger vein image.

[0013] Further, in step 2), the step of image enhancement processing includes:

[0014] 2.1) The normalized finger vein image A is processed by using a limited contrast adaptive histogram equalization algorithm to obtain an image A1;

[0015] 2.2) performing Wiener filtering and median filtering on the image A1 in step 2.1) to obtain image A2;

[0016] 2.3) For each pixel of image A2 in step 2.2), the convolution response of the corresponding 8 direction operators is calculated in a 9×9 window centered at the pixel, and the 8 directions are 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°, and then the maximum convolution response in these 8 directions is used as the new pixel value of the pixel to obtain image A3;

[0017] 2.4) Adaptively thresholding the image A3 in step 2.3) to obtain image A4;

[0018] 2.5) Performing an opening operation on the image A4 in step 2.4) to obtain the final finger vein enhanced image.

[0019] Further, the step 3) specifically includes:

[0020] 3.1) Take the pixel point at the upper left corner of the finger vein image as the origin, the horizontal rightward as the x-axis, and the vertical downward as the y-axis to establish a rectangular coordinate system with size as The rectangle traverses the finger vein image to be identified after step 2), calculates the sum of the grayscale values ​​of the pixels in each rectangular area, and takes the rectangular area corresponding to the largest Sum as the area with the richest feature information. At this time, the center of the rectangular area is (a, b);

[0021] The Sum calculation formula is: Sum = ∑Fgray (m, n)

[0022] In the formula, (m,n) represents the relative coordinates of the pixel point in the rectangular area and the pixel point in the upper left corner of the rectangle. Fgray(m,n) represents the gray value corresponding to the pixel at position (m,n), and Sum represents the sum of the gray values ​​in the rectangular area;

[0023] 3.2) For the region with the richest feature information obtained in step 3.1), horizontal and vertical projections are performed to obtain the horizontal projection vector X and the vertical projection vector Y, which are connected in series to obtain the regional grayscale projection vector q, as follows:

[0024]

[0025] Where, X m Represents the horizontal projection vector of the mth row;

[0026]

[0027] In the formula, X0 is X when m=0 m Value, for Time m Get value;

[0028]

[0029] Where Y n Represents the vertical projection vector of the nth column;

[0030]

[0031] Where, Y0 is Y when n=0 m Value, for Time m Get value;

[0032] q=[X,Y].

[0033] Further, in step 5), the p of each finger vein image in the finger vein database obtained in step 4) is i Calculate the similarity dist(p) with q obtained in step 3) i ,q), take the highest dist(p i ,q) to sort;

[0034] dist(p i ,q)=||p i -q|| ∞

[0035] In the formula, ||.|| ∞ Represents the infinity norm of a vector.

[0036] Further, in step 6), the image numbered as a candidate in the finger vein image of the finger vein database enhanced in step 2) is selected, and the multi-scale HOG features are extracted for matching with the finger vein image to be identified enhanced in step 2). The specific steps include:

[0037] 6.1) Perform trilinear interpolation scaling to obtain finger vein images of four scales, namely w×h,

[0038] 6.2) Extract HOG features from each scale of the finger vein image and concatenate them to obtain the final multi-scale HOG features;

[0039] 6.3) Match the multi-scale HOG features and return the finger vein image number with the highest matching degree as the retrieval result of the finger vein image to be identified, so as to realize the real-time retrieval of large-scale finger vein images.

[0040] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0041] 1. The large-scale finger vein image retrieval method proposed in the present invention uses the sum of the grayscale values ​​of the pixels in the area as a measurement standard when obtaining the area with the richest feature information of the finger vein image to be identified. Not only is the calculation simple, but the grayscale information can also well reflect the characteristics of the finger vein image.

[0042] 2. In the large-scale finger vein image retrieval method proposed in the present invention, the rectangular area is projected horizontally and vertically to obtain the regional grayscale projection vector. The calculation of the regional grayscale projection vector is simple, the dimension is low, and the space occupied is small. It can speed up the retrieval speed when calculating the similarity. At the same time, the projection vector area can fully consider the grayscale and grayscale change information in the horizontal and vertical directions, and well reflect the finger vein characteristics.

[0043] 3. In the large-scale finger vein image retrieval method proposed in the present invention, when obtaining the grayscale projection vector of the large-scale finger vein image area in the database, it moves within the neighborhood of the center point of the area with the richest feature information of the finger vein image to be identified. Compared with traversing the entire finger vein image, the amount of calculation is greatly reduced, and the possible offset phenomenon of the finger vein image is also fully taken into account.

[0044] 4. In the large-scale finger vein image retrieval method proposed in the present invention, the candidate number is selected by using the similarity of the regional grayscale projection vector. Only the finger vein image corresponding to the candidate number needs to be further identified and matched using multi-scale HOG, which greatly reduces the number of times the multi-scale HOG is obtained, greatly improves the retrieval speed, and can achieve real-time performance while maintaining the accuracy of the retrieval.

[0045] 5. The large-scale finger vein image retrieval method proposed in the present invention has the advantages of simple implementation, small amount of calculation, and strong real-time performance, and can be transplanted and deployed on embedded devices in various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Flow chart of the method of the present invention.

[0047] Figure 2 Schematic diagram of the normalized finger vein image in the method of the present invention.

[0048] Figure 3 Schematic diagram of a finger vein image after image enhancement in the method of the present invention.

[0049] Figure 4 It is a schematic diagram of the process of obtaining the regional grayscale projection vector from the finger vein image in the method of the present invention. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0051] like Figures 1 to 4 As shown, this embodiment provides a large-scale finger vein image real-time retrieval method, including the following steps:

[0052] 1) The Shandong University finger vein dataset SDUMLA-HMT Database is a large-scale finger vein database, which contains a total of 3816 finger vein images. The finger vein image to be identified is one of the finger vein images. The finger vein image is grayscale normalized and scale normalized respectively. After scale normalization, the image size is 100×60, as shown in Figure 2 shown.

[0053] 2) Perform image enhancement processing on the normalized finger vein image. The enhanced finger vein image is as follows: Figure 3 As shown, the steps include:

[0054] 2.1) The normalized finger vein image A is processed by using a limited contrast adaptive histogram equalization algorithm to obtain an image A1;

[0055] 2.2) performing Wiener filtering and median filtering on the image A1 in step 2.1) to obtain image A2;

[0056] 2.3) For each pixel of image A2 in step 2.2), the convolution response of the corresponding 8 direction operators is calculated in a 9×9 window centered on the pixel, and the 8 directions are 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°, and then the maximum convolution response in these 8 directions is used as the new pixel value of the pixel to obtain image A3;

[0057] 2.4) Adaptively thresholding the image A3 in step 2.3) to obtain image A4;

[0058] 2.5) Performing an opening operation on the image A4 in step 2.4) to obtain the final finger vein enhanced image.

[0059] 3) Obtain the grayscale projection vector of the region of the finger vein image to be identified and obtain the center point position, such as Figure 4 As shown, the specific steps include:

[0060] 3.1) Traverse the finger vein image to be identified after step 2) with a rectangle of size 50×30, calculate the sum of the grayscale values ​​of the pixels in each rectangular area, and take the rectangular area corresponding to the largest sum as the area with the richest feature information. At this time, the center of the rectangular area is (a, b);

[0061] The Sum calculation formula is: Sum = ∑Fgray (m, n)

[0062] Wherein, (m, n) represents the relative coordinates of the pixel point in the rectangular area and the pixel point in the upper left corner of the rectangle, 0<=m<=50, 0<=n<=30, Fgray(m, n) represents the gray value corresponding to the pixel point, and Sum represents the sum of the gray values ​​in the rectangular area.

[0063] 3.2) For the region with the richest feature information obtained in step 3.1), perform horizontal and vertical projection to obtain the horizontal projection vector X and the vertical projection vector Y, and concatenate them to obtain the regional grayscale projection vector q;

[0064]

[0065] Where, X m Represents the horizontal projection vector of the mth row;

[0066] X=[X0,...,X 29 ]

[0067] In the formula, X0 is X when m=0 m Value, X 29 When m=29, X m Get value;

[0068]

[0069] Where Y n Represents the vertical projection vector of the nth column;

[0070] Y=[Y0,...,Y 49 ]

[0071] Where Y0 is Y when n=0 m Value, Y 49 When n=49, Y m Get value;

[0072] q=[X,Y], the dimension of X is 30, the dimension of Y is 50, and the dimension of q is 80.

[0073] 4) Obtain the finger vein image of the finger vein database after the enhancement processing in step 2), and move and traverse a rectangular area with a size of (50,30) and a center point of (a0,b0) in each image to obtain multiple regional grayscale projection vectors p i , where a-25<=a0<=a+25, b-15<=b0<=b+15;

[0074] 5) The p of each finger vein image in the finger vein database obtained in step 4) is i Calculate the similarity dist(p) with q obtained in step 3) i ,q), take the highest dist(p i ,q) to sort;

[0075] dist(p i ,q)=||p i -q|| ∞

[0076] In the formula, ||.|| ∞ Represents the infinity norm of a vector.

[0077] 6) Take the corresponding number of the finger vein image with a ranking less than the threshold value 0.05*3185 as the candidate, select the image with the corresponding number in the finger vein image of the finger vein database enhanced by step 2), and the finger vein image to be identified enhanced by step 2), respectively, and perform trilinear interpolation scaling to obtain finger vein images of four scales, the four scales are (100,60), (50,60), (100,30), (50,30), respectively. Extract HOG features from the finger vein image of each scale, and concatenate them to obtain the final multi-scale HOG features. Match the multi-scale HOG features, and return the finger vein image number with the highest matching degree as the real-time retrieval result of the finger vein image to be identified, and the experimental process ends.

[0078] The processor of this example is a 2.3GHz Core i5-6200. The retrieval time for the entire large-scale finger vein database is 5.4s, and the single retrieval time is 1.72ms. Compared with the previous retrieval time of 17.54ms using multi-scale HOG features directly, the retrieval speed has been greatly improved, and real-time performance can be achieved.

[0079] The example results show that the large-scale finger vein image real-time retrieval method provided by the present invention greatly improves the retrieval speed and greatly reduces the retrieval time. While taking into account the accuracy, it can realize the real-time retrieval of large-scale finger vein images and is worthy of promotion.

[0080] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principles of the present invention shall be equivalent replacement modes and shall be included in the protection scope of the present invention.

Claims

1. A large-scale finger vein image real-time retrieval method, characterized in that: The following steps are involved: 1) Obtain the finger vein image to be identified and the large-scale finger vein database, normalize the image, and obtain an image size of w×h, where the length is w and the width is h; 2) performing image enhancement processing on the finger vein image normalized in step 1); 3) The finger vein image to be identified after enhancement processing is obtained from step 2) and the size is The rectangle traverses the image, obtains the area with the richest feature information, obtains the center point (a, b) of the area with the richest feature information, and performs horizontal and vertical projection on the area to obtain the regional grayscale projection vector q. The specific steps include: 3.1) Take the pixel point at the upper left corner of the finger vein image as the origin, the horizontal rightward as the x-axis, and the vertical downward as the y-axis to establish a rectangular coordinate system with size as The rectangle traverses the finger vein image to be identified after step 2), calculates the sum of the grayscale values ​​of the pixels in each rectangular area, and takes the rectangular area corresponding to the largest Sum as the area with the richest feature information. At this time, the center of the rectangular area is (a, b); 3.2) For the region with the richest feature information obtained in step 3.1), horizontal and vertical projections are performed to obtain horizontal projection vector X and vertical projection vector Y, which are connected in series to obtain the regional grayscale projection vector q; 4) Take (a, b) in step 3) as the center to obtain the neighborhood range, and the point in the neighborhood range is (a0, b0), where Get the finger vein images of the finger vein database after the enhancement processing in step 2), and in each image, use the size of The rectangular area with the center (a0, b0) is moved and traversed and horizontally and vertically projected to obtain multiple regional grayscale projection vectors p for each image. i ; 5) The p of each finger vein image in the finger vein database obtained in step 4) is i Calculate the similarity with q obtained in step 3), and sort the finger vein images by the value with the highest similarity; 6) According to the sorting result of step 5), the corresponding number of the finger vein image with a ranking less than the threshold threshold is taken as a candidate, and the image numbered as a candidate in the finger vein image of the finger vein database enhanced by step 2) is selected, and the multi-scale HOG features are extracted and matched with the finger vein image to be identified enhanced by step 2), and the finger vein image number with the highest matching degree is returned as the finger vein image real-time retrieval result.

2. A large-scale finger vein image real-time retrieval method according to claim 1, characterized in that: In step 2), the image enhancement process includes: 2.1) The normalized finger vein image A is processed by using a limited contrast adaptive histogram equalization algorithm to obtain an image A1; 2.2) performing Wiener filtering and median filtering on the image A1 in step 2.1) to obtain image A2; 2.3) For each pixel of image A2 in step 2.2), the convolution response of the corresponding 8 direction operators is calculated in a 9×9 window centered at the pixel, and the 8 directions are 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°, and then the maximum convolution response in these 8 directions is used as the new pixel value of the pixel to obtain image A3; 2.4) Adaptively thresholding the image A3 in step 2.3) to obtain image A4; 2.5) Performing an opening operation on the image A4 in step 2.4) to obtain the final finger vein enhanced image.

3. The large-scale finger vein image real-time retrieval method according to claim 1 is characterized in that: In step 3.1), the Sum calculation formula is: Sum = ∑Fgray(m,n) In the formula, (m,n) represents the relative coordinates of the pixel point in the rectangular area and the pixel point in the upper left corner of the rectangle. Represents the grayscale value corresponding to the pixel at position (m,n), and Sum represents the sum of the grayscale values ​​in the rectangular area; In step 3.2), Where, X m Represents the horizontal projection vector of the mth row; In the formula, X0 is X when m=0 m Value, for Time m Get value; Where Y n Represents the vertical projection vector of the nth column; Where, Y0 is Y when n=0 m Value, for Time m Get value; q=[X,Y].

4. The large-scale finger vein image real-time retrieval method according to claim 1, characterized in that: In step 5), the p of each finger vein image in the finger vein database obtained in step 4) is i Calculate the similarity dist(p) with q obtained in step 3) i ,q), take the highest dist(p i ,q) to sort; dist(p i ,q)=||p i -q|| ∞ In the formula, ||.|| ∞ Represents the infinity norm of a vector.

5. The large-scale finger vein image real-time retrieval method according to claim 1, characterized in that: In step 6), the image numbered as a candidate in the finger vein image of the finger vein database enhanced in step 2) is selected, and the multi-scale HOG features are extracted for matching with the finger vein image to be identified enhanced in step 2). The specific steps include: 6.1) Perform trilinear interpolation scaling to obtain finger vein images of four scales, which are: 6.2) Extract HOG features from each scale of the finger vein image and concatenate them to obtain the final multi-scale HOG features; 6.3) Match the multi-scale HOG features and return the finger vein image number with the highest matching degree as the retrieval result of the finger vein image to be identified, so as to realize the real-time retrieval of large-scale finger vein images.

Images