A fingerprint image fusion method based on small-area phase correlation

Abstract

This invention discloses a fingerprint image fusion method based on small-area phase correlation, belonging to the field of image processing technology. The method includes the following steps: S1: preprocessing the acquired local fingerprint image; S2: obtaining the phase information of the preprocessed local fingerprint image; S3: constructing a small-area phase correlation, calculating the similarity between local fingerprint images, and finding the sub-region with the highest phase correlation; S4: fusing the sub-regions with the highest phase correlation in the local fingerprint image. This invention utilizes small-area phase correlation technology to effectively fuse small-area fingerprint images and low-quality fingerprint images, fusing various local regions from the same finger, making the fingerprint image information more complete, and solving the problem of extremely poor feature-based fusion results due to the limited number of feature points in the acquired local fingerprint images.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a fingerprint image fusion method based on small-area phase correlation. Background Technology

[0002] Fingerprints have been used for personal biometric identification for over a century. Due to their immutability and uniqueness, they are widely used in biometric authentication, such as in criminal investigations, where they can quickly assist police in solving cases. However, fingerprints collected at crime scenes are often incomplete and flawed, representing partial fingerprint images. These partial images contain limited effective information, specifically fewer minutiae. Typically, a small fingerprint image contains 3 to 15 minutiae. For minutiae-based matching, at least 7 to 12 pairs of minutiae are needed for a successful match. Furthermore, not all 3 to 15 minutiae in a small area will necessarily correspond one-to-one, and false matches exist, leading to poor matching results. Therefore, police urgently need an efficient and applicable partial fingerprint image fusion technology to ensure the integrity of the suspect's fingerprint image information.

[0003] Traditional feature-based fingerprint image fusion methods primarily utilize feature point information to fuse fingerprints. This method is highly effective for fusing high-quality fingerprint images; however, for small-area or low-quality fingerprint images, the fusion results are poor due to the limited effective information available for fingerprint fusion, i.e., fewer feature points.

[0004] Currently, most fingerprint image fusion methods have poor applicability, meaning they are only suitable for fusion of fingerprint images collected in a specific scenario. Therefore, further improving the fusion of fingerprint images collected in multiple scenarios is of great significance both theoretically and practically. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides a fingerprint image fusion method based on small-area phase correlation.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A fingerprint image fusion method based on small-area phase correlation includes the following steps:

[0008] S1: Perform image segmentation and enhancement processing on the acquired local fingerprint image.

[0009] S2: Calculate the phase information of the local fingerprint image processed in step (1) using Fourier transform: Assume f roi (x,y) is the preprocessed local fingerprint image, then f roiThe Fourier transform of (x,y) is as follows:

[0010]

[0011] Where: x and y are the spatial sampling values ​​of the local fingerprint image; u and v are the frequency sampling values ​​of the local fingerprint image; j is the imaginary unit; F roi (u,v) is f roi The Fourier transform amplitude of (x,y), F roi The integral of (u,v) is a complex-valued function, therefore (1) can be written as:

[0012]

[0013] Where: u and v are the sampled values ​​of the local fingerprint image in the frequency domain; A(u,v) is the amplitude; θ(u,v) is the phase function of the local fingerprint image after Fourier transform; j is the imaginary unit.

[0014] Local fingerprint image f roi Fourier transform F of (x,y) roi The real part of (u,v) corresponds to f roi The projection of (x,y) onto the cosine function, with its imaginary part corresponding to f roi The projection of (x,y) onto the sine function, and the ratio of the two projections, determines the phase function.

[0015] Replacing the cosine and sine functions with orthogonal functions in the orthogonal basis, equation (2) becomes:

[0016]

[0017] Where: η and γ are a set of orthogonal bases; x and y are the sampled values ​​of the local fingerprint image in the spatial domain; u and v are the sampled values ​​of the local fingerprint image in the frequency domain; A(u,v) is the amplitude; θ(u,v) is the phase function of the local fingerprint image after Fourier transform; j is the imaginary unit.

[0018] The phase information of the local fingerprint image is calculated according to equation (3) as f. roi The arctangent of the projection ratio of (x,y) onto two orthogonal functions.

[0019] S3: Then, based on the phase information from step (2), construct the phase correlation, calculate the similarity between local fingerprints, and find the sub-region with the highest phase correlation:

[0020] Assume F roi1 (u,v), F roi2 (u, v) represent the local fingerprint images f. roi1 (x,y),f roi2The Fourier transform of (x,y) then f roi1 (x,y) and f roi2 The phase correlation spectrum C(u,v) of (x,y) in the spatial domain is:

[0021]

[0022] Where u and v are the sampled values ​​of the local fingerprint image in the frequency domain; Phase difference; These are local fingerprint images f roi1 (x,y),f roi2 (x,y) is the phase function after Fourier transform; j is the imaginary unit; F roi2 (u,v) * For F roi2 The complex conjugate of (u,v).

[0023] After performing an inverse Fourier transform on the above equation, the phase correlation function c(x,y) is obtained:

[0024]

[0025] Where: x and y are the sampled values ​​of the local fingerprint image in the spatial domain; u and v are the sampled values ​​of the local fingerprint image in the frequency domain; w and h are the dimensions of the local fingerprint image; c(x,y) is the phase correlation function of two local fingerprint images, i.e., the similarity of the local fingerprint images.

[0026] When the peak value of the phase correlation function c(x,y) is less than 0.1, the local fingerprint image f roi1 Using (x, y) as a reference, firstly, with the x and y axes as axes of symmetry respectively, the local fingerprint image f is analyzed. roi2 The local fingerprint image is mirrored and rotated sequentially (x, y) to obtain the maximum phase correlation, i.e., the peak value of the phase correlation function. The value of the maximum phase correlation is between 0.1 and 1, and the maximum phase correlation is different between different local fingerprint images. At this time, the position corresponding to the peak value of the phase correlation function is the position of the sub-region with the highest phase correlation.

[0027] S4: Merge the sub-regions with the highest phase correlation in the local fingerprint image:

[0028] With local fingerprint image f roi1 Using (x,y) as the reference, let f be the local fingerprint image in S3 that has undergone rotation and has the highest phase correlation. roi2_rot (x,y), translation f roi2_rot (x,y) until f roi2_rot (x,y) and f roi1 Until the (x,y) similar subregions overlap, let the translated local fingerprint image be f. roi2_rot_tra(x,y), the two local fingerprint images are fused according to the following formula:

[0029] f(x,y)=α×f roi1 (x,y)+(1-α)×f roi2_rot_tra (x,y)

[0030] Where: x and y are the sampled values ​​of the local fingerprint image in the spatial domain; α represents the weight; f(x,y) is the fused fingerprint image.

[0031] In a preferred embodiment of the present invention, in step S3, when the peak value of the phase correlation function c(x,y) is less than 0.1, the local fingerprint image f is used. roi1 Using (x,y) as a reference, identify the sub-regions with the highest phase correlation, specifically including:

[0032] SS1: Using the x and y axes as axes of symmetry respectively, analyze the local fingerprint image f... roi2 The x-axis mirror image f can be obtained by mirroring (x,y) along the positive x-axis, positive y-axis, or negative y-axis. roi2_x (x, y), mirror image f along the positive y-axis roi2_+y (x,y) and mirror image f along the negative y-axis roi2_-y (x,y), and calculate the mirror images of the positive x-axis, positive y-axis, and negative y-axis directions respectively, and compare them with f. roi1 The phase correlation of (x,y) is calculated, and the mirror image with the highest phase correlation is retained.

[0033] SS2: Within the range of 0–90°, the mirror image with the highest retained phase correlation is rotated three times with a rotation interval of 30° to obtain three rotated images. The correlation between these three rotated images and f is then calculated. roi1 Phase correlation of (x,y) is used to retain the rotating fingerprint image with the highest phase correlation.

[0034] SS3: Rotate the retained rotating fingerprint image with the highest phase correlation within the range of 0 to 30° at rotation intervals of 2° to obtain 15 new rotating fingerprint images. Calculate the relationship between each of the 15 new rotating fingerprint images and f. roi1 The phase correlation of (x,y) can be used to find the maximum phase correlation of the local fingerprint image. The position corresponding to the peak of the phase correlation function is the position of the sub-region with the highest phase correlation.

[0035] In a preferred embodiment of the present invention, the specific steps of image segmentation and enhancement processing in S1 include: using adaptive threshold segmentation to segment the foreground and background of the acquired local fingerprint image, and then using a Gabor filter to enhance the texture of the segmented local fingerprint image.

[0036] Compared with existing technologies, the advantages of this invention are as follows: This invention fully utilizes the uniqueness of phase, achieving excellent fusion results not only for high-quality fingerprint images but also for small-area and low-quality fingerprint images, effectively solving the problem of poor feature-based fusion results due to the limited number of feature points in the acquired local fingerprint images. Furthermore, the algorithm described in this invention does not require real-time parameter adjustment; instead, it finds the optimal position of phase correlation through the algorithm, thereby achieving the fusion of multiple local fingerprints from the same finger. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the fingerprint image fusion method based on small-area phase correlation of the present invention.

[0038] Figure 2 The original partial fingerprint image collected in this invention: Figure 2 (a) Figure 2 (b) represents the local fingerprint images f respectively. roi1 (x,y) and f roi2 (x,y).

[0039] Figure 3 This is a segmentation diagram of a local fingerprint image acquired in this invention; Figure 3 (a) Figure 3 (b) respectively correspond to Figure 2 (a) Figure 2 (b) is a segmented image.

[0040] Figure 4 This is an enhanced image of a local fingerprint image acquired in this invention; Figure 4 (a) Figure 4 (b) respectively correspond to Figure 3 (a) Figure 3 (b) Enhanced image.

[0041] Figure 5 This is a phase correlation map of a local fingerprint image acquired in this invention.

[0042] Figure 6 This is a map showing the locations of similar sub-regions in the local fingerprint image acquired in this invention: Figure 6 (a) is a local fingerprint image f roi1 Location map of similar subregions in (x,y). Figure 6 (b) is a local fingerprint image f roi2 Location map of similar subregions of (x,y).

[0043] Figure 7 This is a fusion map of sub-region locations in a local fingerprint image acquired in this invention. Detailed Implementation

[0044] To better illustrate the purpose, technical solution, and advantages of the present invention, the present invention will be further described below in conjunction with specific embodiments.

[0045] Example 1

[0046] A fingerprint image fusion method based on small-area phase correlation technology includes the following steps:

[0047] S1: Preprocessing the acquired local fingerprint image: Adaptive threshold segmentation is used to segment the foreground and background of the acquired local fingerprint image. Then, the orientation field and frequency field of the local fingerprint image are calculated, and a 3×3 Gabor filter window is used to enhance the texture of the local fingerprint image. The acquired local fingerprint image is shown below. Figure 2 As shown, Figure 2 (a) Figure 2 (b) represents the local fingerprint images f respectively. roi1 (x,y) and f roi2 (x,y), the segmentation effect is as follows Figure 3 As shown, the ridges and valleys of the local fingerprint image after adaptive threshold segmentation have significantly improved clarity, but some spurs still remain. The enhancement effect is as follows: Figure 4 As shown, from Figure 4 (a) Figure 4 (b) It can be seen that the fingerprint texture is smoother and clearer and the burrs have been removed.

[0048] S2: Obtain the phase information of the preprocessed local fingerprint image: Let f roi (x,y) is the preprocessed local fingerprint image, then f roi The Fourier transform of (x,y) is as follows:

[0049]

[0050] Where: x and y are the spatial sampling values ​​of the local fingerprint image; u and v are the frequency sampling values ​​of the local fingerprint image; j is the imaginary unit; F roi (u,v) is f roi The Fourier transform amplitude of (x,y), F roi The integral of (u,v) is a complex-valued function, therefore it can be written as:

[0051] F roi (u,v)=A(u,v)e jθ(u,v) =A(u,v)(cos(θ(u,v))+jsin(θ(u,v)))

[0052] Where: u and v are the sampled values ​​of the local fingerprint image in the frequency domain; A(u,v) is the amplitude; θ(u,v) is the phase function of the local fingerprint image after Fourier transform; j is the imaginary unit. From the above equation, it can be seen that the local fingerprint image f roi Fourier transform F of (x,y) roi The real part of (u,v) corresponds to f roi The projection of (x,y) onto the cosine function, with its imaginary part corresponding to f roi The projections of (x, y) onto the sine function are such that the ratio of these two projections determines the phase function. To more intuitively understand the phase information, let's replace the orthogonal basis of the cosine and sine functions with other orthogonal functions. Then the above equation can be written as:

[0053]

[0054] Where: η and γ form an orthogonal basis; x and y are the sampled values ​​of the local fingerprint image in the spatial domain; u and v are the sampled values ​​of the local fingerprint image in the frequency domain; A(u,v) is the amplitude; θ(u,v) is the phase function of the local fingerprint image after Fourier transform; and j is the imaginary unit. From the above equation, it can be seen that the phase information of the local fingerprint image is f. roi The arctangent of the projection ratio of (x,y) onto two orthogonal functions.

[0055] S3: Construct small-area phase correlation, calculate the similarity between local fingerprint images, and find the location of the sub-region with the highest phase correlation: Let F roi1 (u,v), F roi2 (u,v) represents the local fingerprint image f. roi1 (x,y),f roi2 The Fourier transform of (x,y) then f roi1 (x,y) and f roi2 The phase correlation spectrum C(u,v) of (x,y) in the spatial domain is:

[0056]

[0057] Where u and v are the sampled values ​​of the local fingerprint image in the frequency domain; Phase difference; These are local fingerprint images f roi1 (x,y),f roi2 (x,y) is the phase function after Fourier transform; j is the imaginary unit; F roi2 (u,v) * For F roi2 The complex conjugate of (u,v) is used to obtain the phase correlation function c(x,y) by performing an inverse Fourier transform on the above equation:

[0058]

[0059] Where: x and y are the sampled values ​​of the local fingerprint image in the spatial domain; u and v are the sampled values ​​of the local fingerprint image in the frequency domain; w and h are the dimensions of the local fingerprint image; c(x,y) is the phase correlation function of two local fingerprint images, i.e., the similarity of the local fingerprint images.

[0060] like Figure 5 As shown, from Figure 5 It can be seen that at position (39, 81), the peak value of the phase correlation function is 0.11052, that is, the phase correlation is 0.11052. At this time, the local fingerprint image f is... roi1 (x,y) and f roi2 This position marker (x, y), i.e. Figure 6 (a) Figure 6 (b) shows the location of the sub-region outlined in red. When the peak value of the phase correlation function c(x,y) is less than 0.1, the local fingerprint image f... roi1 Using (x, y) as a reference, firstly, with the x and y axes as axes of symmetry respectively, the local fingerprint image f is analyzed. roi2 By performing a mirror transformation on (x,y), we can obtain the x-axis mirror image f. roi2_x (x, y), mirror image f along the positive y-axis roi2_+y (x,y) and mirror image f along the negative y-axis roi2_-y (x,y), and calculate the mirror images in these three directions and f respectively. roi1 The phase correlation of (x,y) is calculated, and the mirror image with the highest phase correlation is retained. Then, within the range of 0 to 90°, the retained mirror image is rotated three times with a rotation interval of 30°, resulting in three rotated images. The phase correlation between these three rotated images and f is calculated for each image. roi1 The phase correlation of (x,y) is calculated, and the rotated fingerprint image with the highest phase correlation is retained. Then, the retained rotated fingerprint image with the highest phase correlation is rotated at a rotation interval of 2°, resulting in 15 new rotated fingerprint images. The phase correlation of these 15 fingerprint images with f is then calculated. roi1 The phase correlation of (x,y) can be used to find the maximum phase correlation of the local fingerprint image. The position corresponding to the peak of the phase correlation function is the position of the sub-region with the highest phase correlation.

[0061] S4: Merge the sub-regions with the highest phase correlation in the local fingerprint image to obtain the local fingerprint image f roi1 Using (x,y) as the reference, let f be the local fingerprint image in S3 that has undergone rotation and has the highest phase correlation. roi2_rot (x,y), translation f roi2_rot (x,y) until f roi2_rot (x,y) and f roi1Until the (x,y) similar subregions overlap, let the translated local fingerprint image be f. roi2_rot_tra Based on (x, y), the two local fingerprint images can be fused using the following formula:

[0062] f(x,y)=α×f roi1 (x,y)+(1-α)×f roi2_rot_tra (x,y)

[0063] Where: x and y are the spatial sampling values ​​of the local fingerprint image; α represents the weight; f(x,y) is the fused fingerprint image, such as... Figure 7 As shown, from Figure 7 It can be seen that the invention can effectively fuse local fingerprint images, and the fused fingerprint images do not have any improper connection, and the fingerprint texture information is also richer and more complete.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A small-area phase correlation based method for fusing fingerprint images, characterized in that, It comprises the following steps: S1: image segmentation and enhancement processing are performed on the acquired local fingerprint image; S2: Phase information calculation on the local fingerprint image after step (1) by Fourier transform: Assume f roi (x,y) is the pre-processed local fingerprint image, then the Fourier transform of f roi (x,y) is as follows: where: x, y are the sample values of the local fingerprint image in spatial domain; u, v are the sample values of the local fingerprint image in frequency domain; j is the imaginary unit; F roi (u, v) is the Fourier transform amplitude of f roi (x, y), F roi The integral result of (u, v) is a complex-valued function, so (1) is written as: Wherein: u, v are the sampling values of the local fingerprint image in the frequency domain; A(u, v) is the amplitude; θ(u, v) is the phase function of the local fingerprint image after Fourier transform; j is the imaginary unit; local fingerprint image f roi Fourier transform F of (x,y) roi real part of (u,v) corresponds to f roi projection of (x,y) on cosine function, imaginary part corresponds to f roi projection of (x,y) on sine function, ratio of the two projections determines the phase function; The orthogonal function is replaced with the orthogonal basis of cosine function and sine function, and formula (2) is converted into: Wherein: η and γ are a set of orthogonal bases; x, y are the sampling values of the local fingerprint image in the spatial domain; u, v are the sampling values of the local fingerprint image in the frequency domain; A(u, v) is the amplitude; θ(u, v) is the phase function of the local fingerprint image after Fourier transform; j is the imaginary unit; The phase information of the local fingerprint image is calculated according to formula (3) as f roi the inverse tangent of the ratio of the projections of (x, y) on the two orthogonal functions; S3: then, according to the phase information of step (2), a phase correlation is constructed, the similarity between the local fingerprints is calculated, and the sub-region position with the highest phase correlation is found out: Assume F roi1 (u,v), F roi2 (u,v) are the Fourier transforms of the local fingerprint images f roi1 (x,y), f roi2 (x,y) respectively, then the phase correlation spectrum C(u,v) of f roi1 (x,y) and f roi2 (x,y) in the spatial domain is: wherein: u, v are the sample values of the local fingerprint image in the frequency domain; is the phase difference; are the phase functions of the local fingerprint images f roi1 (x,y), f roi2 (x,y) after Fourier transformation; j is the imaginary unit; F roi2 (u,v) * is the complex conjugate of F roi2 (u,v) After inverse Fourier transform of the above formula, the phase correlation function c(x, y) is obtained: Wherein: x, y are the sampling values of the local fingerprint image in the spatial domain; u, v are the sampling values of the local fingerprint image in the frequency domain; w, h are the size of the local fingerprint image; c(x, y) is the phase correlation function of two local fingerprint images, that is, the similarity of the local fingerprint image; When the phase correlation function peak c(x, y) is lower than 0.1, the local fingerprint image f roi1 (x, y) is taken as the reference, the local fingerprint image f roi2 (x, y) is first respectively mirrored and rotated with the x and y axes as the symmetry axes to obtain the maximum phase correlation of the local fingerprint image, i.e. the phase correlation function peak, the value of the maximum phase correlation being between 0.1 and 1, and the maximum phase correlation being different between different local fingerprint images, at which time the position corresponding to the phase correlation function peak is the position of the sub-region with the highest phase correlation. S4: the sub-region position with the highest phase correlation in the local fingerprint image is fused: Let f roi1 (x,y) be the local fingerprint image in S3 with the maximum correlation in phase by rotation, with respect to the reference (x,y). roi2_rot (x,y) be the local fingerprint image in S3 with the maximum correlation in phase by rotation, with respect to the reference (x,y). roi2_rot (x,y) be the local fingerprint image in S3 with the maximum correlation in phase by rotation, with respect to the reference (x,y). roi2_rot (x,y) be the local fingerprint image in S3 with the maximum correlation in phase by rotation, with respect to the reference (x,y). roi1 (x,y) be the local fingerprint image in S3 with the maximum correlation in phase by rotation, with respect to the reference (x,y). roi2_rot_tra (x,y) be the local fingerprint image in S3 with the maximum correlation in phase by rotation, with respect to the reference (x,y). f(x, y) = a x f roi1 (x, y) + (1 - a) x f roi2_rot_tra (x, y) Wherein: x, y are the sampling values of the local fingerprint image in the spatial domain; α represents the weight; f(x, y) is the fused fingerprint image.

2. The method of claim 1, wherein the method is based on a small-area phase correlation. In the S3, when the phase correlation function peak c(x, y) is lower than 0.1, the local fingerprint image f roi1 (x, y) is taken as the reference, the position of the sub-region with the highest phase correlation is found out, specifically including: SS1: respectively taking x, y axis as the symmetry axis, the local fingerprint image f roi2 (x,y) in x axis, y axis positive direction or y axis negative direction mirror image transformation can get x axis mirror image f roi2_x (x,y), y axis positive direction mirror image f roi2_+y (x,y) and y axis negative direction mirror image f roi2_-y (x,y), and respectively calculate x axis, y axis positive direction, y axis negative direction mirror image respectively with f roi1 (x,y) phase correlation, keep the mirror image with the largest phase correlation; SS2: within a range of 0-90°, performing 3 rotations on the reserved mirror image with the largest phase correlation, with a rotation interval of 30°, to obtain 3 rotated images, and calculating the phase correlation of the 3 rotated images with f roi1 (x,y) respectively, and reserving the rotated fingerprint image with the largest phase correlation. SS3: rotate the reserved rotation fingerprint image with the largest phase correlation in a range of 0-30° with a rotation interval of 2°, 15 new rotation fingerprint images can be obtained, and the phase correlation of the 15 new rotation fingerprint images and f roi1 (x, y) is obtained, that is, the maximum phase correlation of the local fingerprint image is obtained, and a position corresponding to a peak value of the phase correlation function is a position of a sub-region with the highest phase correlation.

3. The method of claim 1, wherein the phase correlation is based on a small area. 3 In the S1, the specific steps of image segmentation and enhancement processing include: using adaptive threshold segmentation to segment the collected local fingerprint image into foreground and background, and then using Gabor filter to perform texture enhancement on the local fingerprint image after segmentation processing.

Images