CN109285204B - Biological key generation method for fusing fingerprint and finger vein bit levels - Google Patents
Biological key generation method for fusing fingerprint and finger vein bit levels Download PDFInfo
- Publication number
- CN109285204B CN109285204B CN201811139551.0A CN201811139551A CN109285204B CN 109285204 B CN109285204 B CN 109285204B CN 201811139551 A CN201811139551 A CN 201811139551A CN 109285204 B CN109285204 B CN 109285204B
- Authority
- CN
- China
- Prior art keywords
- image
- fingerprint
- finger vein
- point
- points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/60—Editing figures and text; Combining figures or text
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/12—Fingerprints or palmprints
- G06V40/1347—Preprocessing; Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/14—Vascular patterns
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention relates to a biological key generation method for fusing fingerprint and finger vein bit levels. The invention discloses a biological key training method, which is characterized in that a biological key extraction matrix is trained by fingerprints and finger vein samples collected in the early stage; fingerprint and finger vein fusion biological key extraction after preprocessing a fingerprint sample and a finger vein sample to be extracted, multiplying the preprocessed fingerprint sample and the preprocessed finger vein sample by a key extraction matrix obtained by biological key training to obtain a fingerprint and finger vein fusion biological key. The invention provides a fingerprint and finger vein fusion biological key extraction method by utilizing the characteristics that the biological characteristics are richer after the fingerprint and finger vein fusion and the two biological characteristics can be simultaneously acquired. The fingerprint and finger vein biological characteristic image blind alignment technology is provided, and the accuracy and stability of subsequent biological key extraction are improved; meanwhile, a feature space machine learning method is provided, the fingerprint and finger vein feature vectors are subjected to bit level fusion, and a more stable and accurate biological key than that of the original feature vector direct extraction method can be extracted.
Description
Technical Field
The invention belongs to the technical field of network space security, and relates to a biological key generation method for bit-level fusion of fingerprints and finger veins.
Background
With the rapid development of internet economy in recent years, biometric-based identity authentication is increasingly popular for network identity authentication because of its advantages of convenient use, no need for users to remember keys with great effort, and the like. However, the widespread use of biometric authentication presents challenges in protecting privacy of biometric information. The current biological characteristic authentication technology needs to store the biological characteristic template of the user at the server, and an attacker can completely restore the biological characteristic information of the user as long as the attacker grasps the template, so that the server is successfully deceived, and the user identity logs in a network to obtain related information and resources. The current biological characteristic identity authentication technology has obvious potential safety hazard in actual use.
In order to increase the security of the identity authentication technology, researchers in the field of the technology have proposed a biometric key technology, that is, a client processes a biometric feature and directly extracts a stable digital sequence as a key, a password, and the like for remote identity authentication. The length of the stable digital sequence generally requires more than 128 bits to achieve a certain key strength. The biological key technology enables the server side not to store the biological characteristic information of the client, and reduces the safety risk of using the biological characteristic.
The current research on the biometric key technology is still in the stage of enhancing the stability and strength of the key, and the key extraction technology has many disadvantages. For example, a chinese invention patent ZL201410074438.4 is a human finger vein biometric key generation method, and ZL201410074388.x is a human fingerprint biometric key generation method, which proposes a technical route for projecting feature parameters of a biometric model to a high-dimensional space to obtain a stable biometric key. Compared with the prior art, the technical route has obvious improvement in key stability and key strength, but for the key authentication environment with high stability requirement, the stability and strength of extracting the biological key by the technical route still need to be further improved.
Disclosure of Invention
The invention aims to provide a biometric key generation method for fusing a fingerprint and a finger vein bit level.
The invention comprises biological key training and fingerprint and finger vein fusion biological key extraction; training a biological key to obtain a biological key extraction matrix through fingerprint and finger vein samples collected in the early stage; fingerprint and finger vein fusion biological key extraction, after preprocessing a fingerprint sample and a finger vein sample to be extracted, multiplying the preprocessed fingerprint sample and the preprocessed finger vein sample by a key extraction matrix obtained by biological key training to obtain a fingerprint and finger vein fusion biological key; the method is characterized in that:
the method comprises the following specific steps:
step one, training a finger vein fusion biological key, which comprises the following specific steps:
firstly, a user respectively carries out sample collection on the same finger fingerprint and the finger vein for multiple times to obtain more than 3 fingerprint gray level images and finger vein gray level images, the fingerprint gray level images are uniformly scaled to 354 multiplied by 354 pixel size, the finger vein gray level images are scaled to 256 multiplied by 64 or 256 multiplied by 256 pixel size, and the fingerprint images obtained at this stage are marked as a first fingerprint image and a first finger vein image;
secondly, respectively carrying out equalization, convergence, smoothing, enhancement, binarization and refinement on the first fingerprint image and the first finger vein image obtained in the first step to obtain a preprocessed second fingerprint image and a preprocessed second finger vein image;
the first fingerprint image and the first finger vein image are respectively preprocessed to obtain preprocessed images, the preprocessing processes adopted by the two types of images are basically the same, the difference is only in the value difference of a plurality of parameters, and the preprocessing processes of the first fingerprint image and the first finger vein image are uniformly expressed as follows:
1) Respectively carrying out equalization processing on the first fingerprint image and the first finger vein image; a histogram equalization method, which is a common method in the field of image processing; the histogram equalization formula is shown in the following table:
wherein x represents a gray scale value between 0 and 255; f (x) represents that the gray value of the image is adjusted to f (x) at the point with the gray value of x; d max ＝255，A 0 Is the total number of pixels of the image, H i The number of points with a gray scale value of i in the image is shown;
2) Respectively carrying out convergence processing on the equalized first fingerprint image and the equalized first finger vein image obtained in the step 1); carrying out Gaussian filtering processing, namely image convergence processing, on the image by using a two-dimensional discrete Gaussian template operator; the method is a general method in the field of image processing; a plurality of sets of two-dimensional discrete Gaussian template operators G can be taken as examples
The computational formula is found in the following:
wherein: f (x, y) represents the gray value corresponding to the point with the coordinate (x, y) in the image after the equalization processing;
3) Respectively smoothing the converged first fingerprint image and the converged first finger vein image obtained in the step 2); smoothing the image by using a smooth template operator; the method is a general method in the field of image processing; the smooth template operator T may take multiple sets, taken as an example
The calculation formula is shown as formula (3):
wherein f' (x, y) represents the gray value corresponding to the point with the coordinate (x, y) in the fingerprint image and the finger vein image after smoothing;
4) Respectively carrying out image enhancement processing on the first fingerprint image and the first finger vein image which are obtained in the step 3) and subjected to the smoothing processing; carrying out image enhancement processing on the fingerprint image and the finger vein image after smoothing in the step 3) by using a Gabor function template; the method is a general method in the field of image processing; the Gabor function template may take multiple sets, taken as an example:
determining the tangential direction (the direction perpendicular to the current direction) of a point with the coordinates (x, y) by using the value of VMAP [ x, y ], and taking 7 points adjacent to the point (x, y) in the tangential direction, wherein each point is represented by f (x, y); f (x, y) represents the gray value corresponding to the point with the coordinate of (x, y) in the fingerprint image and the finger vein image; f (x, y) and Hw are subjected to convolution operation, and the operation result is assigned to f1 (x, y); f1 (x, y) is the gray value corresponding to the (x, y) point after the smoothed fingerprint image and the finger vein image are enhanced in the tangential direction;
determining the normal direction (the tangential direction rotates clockwise by 90 degrees and is the normal direction) of a point with the coordinate of (x, y) by using the value of VMAP [ x, y ], taking 7 f1 (x, y) points adjacent to the (x, y) point in the normal direction, performing convolution operation on the points and Vw, and assigning an operation result to f2 (x, y); f2 (x, y) is the gray value corresponding to the (x, y) point after the smoothed fingerprint image and the finger vein image are enhanced in the tangential line and normal line directions;
5) Respectively carrying out binarization processing on the enhanced first fingerprint image and the enhanced first finger vein image obtained in the step 4); the binarization processing is a general algorithm in the field of fingerprint and finger vein identification, and the binarization can be realized by directly calling functions such as im2bw () in matlab and the like;
6) Respectively thinning the first fingerprint image and the first finger vein image which are obtained in the step 5) and subjected to binarization processing; the image processing field has a plurality of mature image thinning methods which can be used;
finishing the steps 1) -6), and obtaining a second fingerprint image and a second finger vein image after the pretreatment is finished;
thirdly, performing blind alignment operation on the second fingerprint image and the second finger vein image to obtain a third fingerprint image and a third finger vein image; although the sampling object is the same fingerprint, the sampling process has the problems of offset, rotation and the like, so that the images sampled for multiple times are not aligned, and the alignment operation is to perform the operations of offset, rotation and the like on the images sampled for multiple times, so that the positions of the acquired objects in the images are basically consistent; blind alignment means that the alignment process is completed only by using the self characteristic information of the current image, and the alignment process is unrelated to the rest images without referring to the images;
the specific process of blind alignment of the fingerprint images comprises the following steps:
(1) Positioning a central point of the fingerprint image; the fingerprint image singular point positioning process is a general method in the field of fingerprint identification, and is generally determined by calculating Poincare formula, assuming possible singular point coordinates as (i, j), calculating Poincare formula (4), and (6):
where θ is the radian, O 'is the direction field (the direction field calculation is a general method in the art), O' (i + ε cos θ, j + ε sin θ) represents the direction field value after the change of the slight radian around the point (i, j),
the direction field difference value of the radian around the point (i, j) after slight change is represented, d delta represents the direction field change trend of the point (i, j) on the radian theta, poincare (i, j) integrates the direction field difference around the point (i, j), and then divides the integration by 2 pi, namely the average direction field difference around the point (i, j);
order to
Is determined as the center point of the fingerprint image; if a plurality of center points exist, taking the mass centers of the center points; if no central point exists, discarding the image, wherein the image cannot be used as a sample for training and testing;
(2) Translating the fingerprint image in the picture along the x-axis direction and the y-axis direction to ensure that the center point of the fingerprint image is superposed with the center point of the picture; the picture is an area with the same size as the fingerprint image, and is a rectangle with the size of 354 multiplied by 354 pixels, the central point of the area is positioned at the origin of a Cartesian coordinate system, and the length and the width are respectively parallel to the y axis and the x axis.
(3) Taking the center point of the image, namely the center point of the fingerprint as the center of a circle, ml pixels as the radius to make a circle, wherein the specific value of ml is preferably that the edge of the circle can vertically cut fingerprint lines, the cutting point is marked as z, and generally, ml takes 15-30 pixels and is determined according to the actual condition; if no vertical cutting point exists, selecting the cutting point closest to the vertical cutting point and recording as z;
(4) Connecting the center point of the fingerprint with the point z, taking a ray, rotating the fingerprint image by taking the center point of the fingerprint as a circle center, and coinciding the ray made by the center point of the fingerprint and the point z with the y axis as a rotation result; the image rotation method is a general method in the image processing field;
finishing the blind alignment processing of the fingerprint images; the fingerprint image blind alignment process is based on the following observations: a region is arranged near the center point of the fingerprint, the center point is used for making rays with the point of the region, and the direction of the rays is approximately parallel to the direction of the lines;
in the process of collecting the finger veins, as a bayonet is generally preset in the collecting equipment, fingers can prop against the top end of the bayonet, so that the positions of the collected finger veins in the length direction are basically fixed without alignment; in the width direction, as the fingers can move left and right in the acquisition process, the acquired images still have the possibility of being misaligned; accordingly, the blind alignment process of the finger vein images mainly considers the alignment in the width direction of the finger vein images;
the specific process is as follows:
(1) performing edge detection on the finger vein image, and extracting a finger vein edge image;
(2) the finger vein edge image is a binary image, has an upper edge and a lower edge, can also be a left edge and a right edge, and the view image placement condition is determined, and a linear regression operation is performed on a point set of the two edges to obtain two straight lines;
(3) placing the finger vein image in a Cartesian coordinate system, placing one vertex of an image rectangle at the origin of the coordinate system, wherein the length direction is parallel to the x axis, the width direction is parallel to the y axis, and the whole image is in the 1 st quadrant;
(4) calculating the center line of the two straight lines, wherein the calculation method comprises the following steps:
suppose that the coordinate equations of the two straight lines are y 1 ＝a1·x+b1，y 2 = a2 · x + b2, calculation
y 3 Is the middle line of two straight lines, wherein a1, b1, a2 and b2 are real numbers, and y1 and y2 are straight line equations in a Cartesian coordinate system;
(5) translating and rotating the finger vein image to enable y3 and the straight line
Overlapping; the specific method can firstly translate the finger vein image to enable y 3 ,y 4 Is coincident with the midpoint of the finger vein image, and then the finger vein image is rotated so that y 3 ,y 4 The two straight lines are completely overlapped; the image translation and rotation method is a common method in the field;
at this point, the blind alignment processing of the vein image is finished;
fourthly, extracting characteristic points from the third fingerprint image and the third finger vein image, recording characteristic point information, and sequentially arranging the characteristic point information into characteristic vectors; the characteristic points generally comprise end points, cross points and the like, pseudo end points and cross points are generally removed in the operation process, and coordinate values of the characteristic points are recorded from top to bottom and from left to right by using a two-dimensional Cartesian coordinate system to form characteristic vectors;
the endpoint extraction method comprises the following steps: scanning 8 points around the points in the thinned third fingerprint image and the third finger vein image, and if the sum of absolute values of differences of all two adjacent points of the 8 points is 2 multiplied by 255, the points are end points;
the method for extracting the cross points comprises the following steps: scanning 8 points around the points in the thinned third fingerprint image and the third finger vein image, and if the sum of absolute values of differences of all two adjacent points of the 8 points is 6 multiplied by 255, the points are cross points;
the method for extracting the end points and the cross points is a general method in the field of fingerprint image processing;
fifthly, machine learning is carried out on the feature vector set obtained in the fourth step to obtain a fingerprint and finger vein fusion feature learning matrix W 1 I.e. a biometric key extraction matrix; fingerprint and finger vein fusion characteristic learning matrix W 1 The sequentially spliced fingerprint and finger vein features are subjected to fusion calculation on a bit level, and the fused features are more stable than the original features;
collecting fingerprints and finger vein samples of a plurality of fingers of different users, repeatedly collecting a plurality of samples by the same finger, and obtaining a characteristic vector set after the four steps of operation; the set is divided into two categories, one category is that of the same finger of the user, and the other category is that of the non-user or the non-current finger and is called as a positive and negative sample set;
with M = [ M = 1 ,M 2 ]Representing positive and negative sample sets participating in the training, M i ＝[x i1 ,x i2 ,...,x iL ]I ∈ {1,2} represents a set of class i samples, i =1 is a positive sample, i =2 is a negative sample; x is the number of ir ∈R d ,1≤i≤2,1≤r≤L，x ir Is a one-dimensional column vector, is a characteristic vector after fingerprint finger vein splicing, and is transposed to obtain a one-dimensional column vector x ir ，x ir Length d, R d Representing a d-dimensional real number domain, wherein L represents a sample characteristic vector obtained by sampling for L times in a sample set of the same finger, namely L column vectors;
now, according to the characteristics of the two types of samples, a fingerprint and finger vein fusion characteristic learning matrix W is trained 1 ∈R d×dz To obtain the formula (7):
wherein
For the positive sample mean of the training samples,
is the negative sample mean of the training sample; j is a cost function and reflects the learning matrix W of the fingerprint and finger vein fusion characteristics of the training sample 1 Calculating the distance difference between the projected image and the positive and negative sample set mean value by using Euclidean distance;
order:
solving matrix (H) 1 -H 2 ) Obtaining a fingerprint and finger vein fusion characteristic learning matrix W by the characteristic value and the characteristic vector of the fingerprint and the finger vein 1 W 1 I.e. (H) 1 -H 2 ) w = λ w; w is a matrix (H) 1 -H 2 ) λ is a eigenvalue;
{w 1 ,w 2 ,...,w dz is an eigenvector corresponding to the eigenvalue { lambda } 1 ,λ 2 ,...,λ dz In which λ is 1 ≥λ 2 ≥...≥λ dz ≧ 0, the eigenvectors with eigenvalues less than 0 are not included in the matrix W 1 The structure of (1); w 1 The fingerprint and finger vein fusion characteristic learning matrix is obtained;
at this point, the fingerprint and finger vein fusion feature learning part is completed to obtain a fingerprint and finger vein fusion feature learning matrix W 1 ；
Step two, fingerprint and finger vein fusion biological key extraction, which comprises the following specific steps:
step 1, collecting a fingerprint and a finger vein image by a user;
2, extracting a gray level image of the fingerprint and the finger vein image, wherein a color image can be used, and the color image is represented by a three-channel gray level image; carrying out equalization, convergence, smoothing, enhancement, binaryzation and thinning on the obtained fingerprint and finger vein images to obtain a thinned image after preprocessing the fingerprint and finger vein images;
step 3, performing blind alignment operation on the fingerprint and the finger vein refined image to obtain an image after the fingerprint and the finger vein are blindly aligned;
step 4, extracting characteristic points from the image after the blind alignment of the fingerprint and the finger vein, wherein the characteristic points generally comprise end points, cross points and the like, and recording characteristic point information; the images after the blind alignment of the fingerprint and the finger vein are respectively placed in a two-dimensional Cartesian coordinate system, the coordinate values of the characteristic points are recorded from top to bottom and from left to right, and the coordinate values of each characteristic point are sequentially placed in a characteristic vector to form a fingerprint and finger vein fusion characteristic vector x t ；
Step 5, learning a matrix W by using the fused features of the trained fingerprints and the finger veins of the first part 1 Left after rotationObtaining the fingerprint finger vein spliced eigenvector x by multiplying the fourth step t I.e. W 1 T ·x t D is obtained z Dimension fused feature vector x tz ；
Step 6, for x tz Each dimension component of (a) is subjected to a chessboard method operation, and the feature vector is further stabilized to
The chessboard method operation process is shown as formula (8):
Λ(x)＝k,(D+1)·k＜x tzi ≤(D+1)·k+D,(k＝0,1,···) (8)；
wherein D is the size of the grid of the chessboard method, a positive number is taken, a specific value can be selected by a user according to experience, the value of Λ (x) is generally between 0 and 63, and x is tz i is x tz Is quantized to integer values; Λ (x) is x tzi The quantized value is the closest x in the checkerboard tzi Coordinate values of the grid of points and the origin of coordinates;
step 7, taking the vector of the calculation result of the step 6
The first n components of (1), n can be 16, 32, 64, etc., and is typically a power of 2, as the case may be
The number of effective characteristic components and the requirement of the biological secret key strength; splicing the n components back and forth to form a fingerprint and finger vein fusion biological key; if n is 64, each component takes a value of 0-64, 4-bit key calculation can be formed, and a result vector
The first n components of (a) may form a 256-bit key sequence;
thus, the fingerprint and finger vein fusion biological key is obtained.
The invention provides a fingerprint and finger vein fusion biological key extraction method by utilizing the characteristics that the biological characteristics are richer after the fingerprint and finger vein fusion and the two biological characteristics can be simultaneously acquired. The method provides a blind alignment technology for fingerprint and finger vein biological characteristic images, supports blind alignment under the condition that a single fingerprint and finger vein image have no reference image, and improves the accuracy and stability of subsequent biological key extraction. The invention provides a feature space machine learning method on the basis of the blind alignment technology of fingerprints and finger veins, performs bit-level fusion on the fingerprint and finger vein feature vectors, and can extract a more stable and accurate biological key than the original feature vector direct extraction method.
Drawings
FIG. 1 is a flow chart of the present invention;
FIGS. 2a and 2b are a fingerprint grayscale image and a finger vein grayscale image in an embodiment;
FIGS. 3a and 3b are the detailed images of the fingerprint and finger vein image after preprocessing in the embodiment;
FIG. 4 is a schematic diagram of a blind alignment process of fingerprint images in an embodiment;
fig. 5 is a schematic diagram of a blind alignment process of finger vein images in the embodiment.
Detailed Description
As shown in fig. 1, a biometric key generation method based on bit level fusion of a fingerprint and a finger vein includes biometric key training and extraction of a fingerprint and finger vein fusion biometric key. Training a biological key to obtain a biological key extraction matrix through fingerprint and finger vein samples collected in the early stage; fingerprint and finger vein fusion biological key extraction the fingerprint and finger vein samples to be extracted are preprocessed and then multiplied by a key extraction matrix obtained by biological key training to obtain the fingerprint and finger vein fusion biological key.
The method comprises the following specific steps:
step one, training a finger vein fusion biological key, which comprises the following specific steps:
firstly, a user respectively carries out sample collection on the same finger fingerprint and the finger vein for multiple times to obtain more than 3 fingerprint gray level images and finger vein gray level images, the fingerprint gray level images are uniformly scaled to 354 multiplied by 354 pixel size, the finger vein gray level images are scaled to 256 multiplied by 64 or 256 multiplied by 256 pixel size, and the fingerprint images obtained at this stage are marked as a first fingerprint image and a first finger vein image, as shown in fig. 2a and 2 b.
And secondly, respectively carrying out equalization, convergence, smoothing, enhancement, binarization and thinning on the first fingerprint image and the first finger vein image obtained in the first step to obtain a preprocessed second fingerprint image and a preprocessed second finger vein image, as shown in fig. 3a and 3 b.
The first fingerprint image and the first finger vein image are respectively preprocessed to obtain preprocessed images, the preprocessing processes adopted by the two images are basically the same, the difference is only in value difference of a plurality of parameters, and the preprocessing processes of the first fingerprint image and the first finger vein image are uniformly expressed as follows:
1) And respectively carrying out equalization processing on the first fingerprint image and the first finger vein image. The histogram equalization method is used, and the method is a common method in the field of image processing. The histogram equalization formula is shown in the following table:
where x represents a gray scale value between 0 and 255. f (x) represents that the gray value of the image is adjusted to f (x) at the point with the gray value of x; d max ＝255，A 0 Is the total number of pixels of the image, H i The number of points in the image with a gray scale value i.
2) And respectively carrying out convergence processing on the equalized first fingerprint image and the equalized first finger vein image obtained in the step 1). And performing Gaussian filtering processing, namely image convergence processing, on the image by using a two-dimensional discrete Gaussian template operator. This method is a common method in the field of image processing. A plurality of sets of two-dimensional discrete Gaussian template operators G can be taken as examples
The computational formula is found in the following:
wherein: f (x, y) represents the gray scale value corresponding to the point with coordinates (x, y) in the image after the equalization processing.
3) And respectively carrying out smoothing treatment on the converged first fingerprint image and the converged first finger vein image obtained in the step 2). And smoothing the image by using a smoothing template operator. This method is a common method in the field of image processing. Smooth template operator T may take multiple sets, taken as an example
The calculation formula is shown in formula (3):
where f' (x, y) represents the grayscale values corresponding to the points with coordinates (x, y) in the smoothed fingerprint image and finger vein image.
4) And respectively carrying out image enhancement processing on the first fingerprint image and the first finger vein image which are obtained in the step 3) after the smoothing processing. And 3) carrying out image enhancement processing on the fingerprint image and the finger vein image after smoothing in the step 3) by using a Gabor function template. This method is a common method in the field of image processing. The Gabor function template may take multiple sets, taken as an example:
The tangential direction (direction perpendicular to the current direction) of the point whose coordinates are (x, y) is determined by the value of VMAP [ x, y ], and 7 points adjacent to the point (x, y) in the tangential direction are taken, each point being represented by f (x, y). f (x, y) represents the gray value corresponding to the point with the coordinate (x, y) in the fingerprint image and the finger vein image. f (x, y) and Hw are subjected to convolution operation, and the operation result is assigned to f1 (x, y). f1 And (x, y) is the gray value corresponding to the (x, y) point after the smoothed fingerprint image and the finger vein image are enhanced in the tangential direction.
And (3) determining the normal direction of a point with coordinates (x, y) by using the value of VMAP [ x, y ] (the normal direction is obtained by clockwise rotating the tangent direction by 90 degrees), taking 7 f1 (x, y) points adjacent to the (x, y) point in the normal direction, performing convolution operation on the points and Vw, and assigning the operation result to f2 (x, y). f2 And (x, y) is the gray value corresponding to the (x, y) point after the fingerprint image and the finger vein image are smoothed and enhanced in the tangential line and normal line directions.
5) And respectively carrying out binarization processing on the enhanced first fingerprint image and the enhanced first finger vein image obtained in the step 4). The binarization processing is a general algorithm in the field of fingerprint and finger vein identification, and the binarization can be realized by directly calling functions such as im2bw () in matlab and the like.
6) And respectively thinning the first fingerprint image and the first finger vein image which are obtained in the step 5) and subjected to the binarization processing. The image processing field has a plurality of mature image thinning methods which can be used.
And finishing the steps 1) to 6), and obtaining a second fingerprint image and a second finger vein image after the preprocessing is finished.
Thirdly, performing blind alignment operation on the second fingerprint image and the second finger vein image to obtain a third fingerprint image and a third finger vein image; although the sampled object is the same fingerprint, the fingerprint and the finger vein image sampled for a plurality of times before and after are misaligned, namely, the images sampled for a plurality of times are misaligned, rotated and the like in the sampling process, so that the positions of the collected objects in the images are basically consistent. Blind alignment means that the alignment process is completed only by using the feature information of the current image, and is independent of the remaining images, and no reference image is needed, as shown in fig. 4.
The specific process of blind alignment of the fingerprint images comprises the following steps:
(1) Positioning a central point of the fingerprint image; the fingerprint image singular point positioning process is a general method in the field of fingerprint identification, and is generally determined by calculating Poincare formula, assuming possible singular point coordinates as (i, j), calculating Poincare formula (4), and (6):
where θ is the radian, O 'is the direction field (the direction field calculation is a general method in the art), O' (i + ε cos θ, j + ε sin θ) represents the direction field value after the change of the slight radian around the point (i, j),
the method is characterized by comprising the following steps of representing the directional field difference value of a slightly changed radian around a point (i, j), d delta represents the directional field change trend of the point (i, j) on theta radian, and Poincare (i, j) integrates the directional field difference around the point (i, j) for one circle, and then divides the integration by 2 pi, namely the average directional field difference around the point (i, j).
Order to
Is determined as the center point of the fingerprint image. If a plurality of center points exist, taking the mass centers of the center points; if no central point exists, discarding the image, wherein the image cannot be used as a sample for training and testing;
(5) Translating the fingerprint image in the picture along the x-axis direction and the y-axis direction to ensure that the center point of the fingerprint image is superposed with the center point of the picture; the picture refers to a region with the same size as the fingerprint image, such as a rectangle with 354 × 354 pixels, the center point of the region is located at the origin of the cartesian coordinate system, and the length and the width are parallel to the y axis and the x axis respectively;
(6) Taking the center point of the image, namely the center point of the fingerprint as the center of a circle, ml pixels as the radius to make a circle, wherein the specific value of ml is that the edge of the circle can vertically cut fingerprint lines, the cutting point is marked as z, and generally, ml takes 15-30 pixels and can be determined according to the actual situation; if no vertical cutting point exists, selecting the cutting point closest to the vertical cutting point and recording as z;
(7) Connecting the center point of the fingerprint with the point z, taking a ray, rotating the fingerprint image by taking the center point of the fingerprint as a circle center, and enabling the ray made by the center point of the fingerprint and the point z to coincide with the y axis as a rotating result, wherein the rotating result is shown in figure 4; the image rotation method is a general method in the image processing field;
and finishing the blind alignment processing of the fingerprint image. The fingerprint image blind alignment process is based on the following observations: an area is arranged near the center point of the fingerprint, the center point is used for taking rays from the point of the area, and the direction of the rays is approximately parallel to the direction of the lines.
In the process of collecting the finger veins, as a bayonet is generally preset in the collecting equipment, fingers can prop against the top end of the bayonet, so that the positions of the collected finger veins in the length direction are basically fixed without alignment; in the width direction, as the fingers can move left and right in the acquisition process, the acquired images still have the possibility of being misaligned; accordingly, the blind alignment process for the finger vein image mainly considers the alignment in the width direction of the finger vein image.
As shown in fig. 5, the specific process is as follows:
(5) carrying out edge detection on the finger vein image, and extracting the finger vein edge image;
(6) the finger vein edge image is a binary image and is provided with an upper edge and a lower edge (or a left edge and a right edge, and the view image placement condition is determined), and a linear regression operation is performed on a point set of the two edges to obtain two straight lines;
(7) placing the finger vein image in a Cartesian coordinate system, placing a certain vertex of an image rectangle at the origin of the coordinate system, wherein the length direction is parallel to the x axis, the width direction is parallel to the y axis, and the whole image is in the 1 st quadrant;
(8) calculating the central line of the two straight lines, wherein the calculation method comprises the following steps:
suppose that the coordinate equations of the two straight lines are y 1 ＝a1·x+b1，y 2 = a2 · x + b2, calculation
y 3 Is the middle line of two straight lines, wherein a1, b1, a2 and b2 are real numbers, and y1 and y2 are straight line equations in a Cartesian coordinate system.
(5) Translating and rotating the finger vein image so that y 3 And a straight line
Overlapping; the specific method can firstly translate the finger vein image to enable y 3 ,y 4 Is coincident with the midpoint of the finger vein image, and then the finger vein image is rotated so that y 3 ,y 4 The two straight lines coincide completely. The image translation and rotation methods are common methods in the art.
So far, the blind alignment processing of the vein image is finished.
Fourthly, extracting characteristic points from the third fingerprint image and the third finger vein image, recording characteristic point information, and sequentially arranging the characteristic point information into characteristic vectors; the feature points generally comprise end points, cross points and the like, pseudo end points and cross points are generally removed in the operation process, and coordinate values of the feature points are recorded from top to bottom and from left to right by using a two-dimensional Cartesian coordinate system to form feature vectors.
The endpoint extraction method comprises the following steps: and scanning 8 points around the points in the refined third fingerprint image and the refined third finger vein image, and if the sum of the absolute values of the differences of all the adjacent two points of the 8 points is 2 multiplied by 255, the points are end points.
The method for extracting the cross points comprises the following steps: and scanning 8 points around the points in the refined third fingerprint image and the refined third finger vein image, wherein if the sum of the absolute values of the differences of all the adjacent two points of the 8 points is 6 multiplied by 255, the points are cross points.
The method for extracting the end points and the cross points is a general method in the field of fingerprint image processing.
Fifthly, performing machine learning on the feature vector set obtained in the fourth step to obtain fingerprints,Vena digitalis fusion feature learning matrix W 1 I.e. the biometric key extraction matrix. Fingerprint and finger vein fusion characteristic learning matrix W 1 And performing fusion calculation on the sequentially spliced fingerprint and finger vein features on a bit level, wherein the fused features are more stable than the original features.
Collecting fingerprints and finger vein samples of a plurality of fingers of different users, repeatedly collecting a plurality of samples by the same finger, and obtaining a characteristic vector set after the four steps of operation; the set is divided into two categories, one category is that of the same finger of the user, and the other category is that of the non-user or the non-current finger, and is called as a positive and negative sample set.
With M = [ M = 1 ,M 2 ]Representing positive and negative sample sets participating in the training, M i ＝[x i1 ,x i2 ,...,x iL ]I ∈ {1,2} represents a set of class i samples, i =1 is a positive sample, i =2 is a negative sample; x is the number of ir ∈R d ,1≤i≤2,1≤r≤L，x ir Is a one-dimensional column vector, is a characteristic vector after fingerprint finger vein splicing, and is transposed to obtain a one-dimensional column vector x ir ，x ir Length d, R d And L represents a sample characteristic vector obtained by sampling for L times in a sample set of the same finger, namely L column vectors.
Now, according to the characteristics of the two types of samples, a fingerprint and finger vein fusion characteristic learning matrix W is trained 1 ∈R d×dz Obtaining the formula (7):
wherein
For the positive sample mean of the training samples,
is the negative sample mean of the training samples. J is a cost function and reflects the learning matrix W of the fingerprint and finger vein fusion characteristics of the training sample 1 Post-projection and positive and negative sample setAnd calculating the distance difference between the combined mean values by using Euclidean distance.
Order:
solving matrix (H) 1 -H 2 ) Obtaining a learning matrix W of the fingerprint and finger vein fusion characteristics 1 W 1 I.e. (H) 1 -H 2 ) w = λ w; w is a matrix (H) 1 -H 2 ) λ is the eigenvalue.
{w 1 ,w 2 ,...,w dz Is an eigenvector corresponding to the eigenvalue { lambda } 1 ,λ 2 ,...,λ dz In which λ is 1 ≥λ 2 ≥...≥λ dz ≧ 0, the eigenvectors with eigenvalues less than 0 are not included in the matrix W 1 The structure of (1). W 1 Namely the fingerprint and finger vein fusion characteristic learning matrix.
At this point, the fingerprint and finger vein fusion feature learning part is completed to obtain a fingerprint and finger vein fusion feature learning matrix W 1 。
Step two, fingerprint and finger vein fusion biological key extraction, which comprises the following specific steps:
step 1, a user collects a fingerprint and a finger vein image.
And 2, extracting a gray level image (a color image can also be used, and the color image can be represented by a three-channel gray level image), and carrying out equalization, convergence, smoothing, enhancement, binarization and thinning on the obtained fingerprint and finger vein image to obtain a thinned image after preprocessing the fingerprint and finger vein image.
And 3, carrying out blind alignment operation on the fingerprint and the finger vein thinned image to obtain an image after the fingerprint and the finger vein are in blind alignment.
And 4, extracting feature points from the image after blind alignment of the fingerprint and the finger vein, wherein the feature points generally comprise end points, cross points and the like, and recording feature point information. The images after the fingerprint and the finger vein are blindly aligned are respectively arranged in a two-dimensional Cartesian coordinate system,recording coordinate values of the feature points from top to bottom and from left to right, sequentially placing the coordinate values of the feature points into the feature vector to form a feature vector x integrating the fingerprint and the finger vein t 。
Step 5, learning a matrix W by using the first part trained fingerprint and finger vein fusion characteristics 1 And after transposition, left-multiplying by the feature vector x obtained in the fourth step after fingerprint finger vein splicing t I.e. W 1 T ·x t D is obtained z Dimension fused feature vector x tz 。
Step 6, for x tz Each dimension component of (a) is subjected to a chessboard method operation, and the feature vector is further stabilized to
The chessboard method operation process is as follows:
Λ(x)＝k,(D+1)·k＜x tzi ≤(D+1)·k+D,(k＝0,1,···) (8)；
wherein D is the size of the grid of the chessboard method, a positive number is taken, a specific value can be selected by a user according to experience, the value of Λ (x) is generally between 0 and 63, and x is tzi Is x tz Is quantized to an integer value. Λ (x) is x tzi The quantized value is the closest x in the checkerboard tzi Point and coordinate value of the grid of origin of coordinates.
Step 7, taking the vector of the calculation result of the step 6
The first n components of (1), n can be 16, 32, 64, etc., and is typically a power of 2, as the case may be
The number of valid feature components and the biometric key strength requirement. And splicing the n components back and forth to form the fingerprint and finger vein fusion biological key. If n is 64, each component takes a value of 0-64, 4-bit key calculation can be formed, and a result vector
The first n components of (a) may form a 256bit key sequence.
Thus, the fingerprint and finger vein fusion biological key is obtained.
The invention extracts the biological key by fusing the fingerprint characteristic and the finger vein characteristic in the bit level, the two types of biological characteristics can expand the effective biological characteristic space and extract more stable biological characteristics, the key length is increased, and the bit level fusion is more favorable for extracting the key sequence presented in the bit form. Compared with a single fingerprint and finger vein biological key extraction method, the method can obtain a more stable and higher-strength biological key, the key extraction accuracy rate can be more than 96% under the user cooperation condition, and the key length can reach 256 bits.
Claims (2)
1. A fingerprint and finger vein bit level fused biological key generation method comprises biological key training and fingerprint and finger vein fused biological key extraction; training a biological key to obtain a biological key extraction matrix through fingerprint and finger vein samples collected in the early stage; fingerprint and finger vein fusion biological key extraction, after preprocessing a fingerprint sample and a finger vein sample to be extracted, multiplying the preprocessed fingerprint sample and the preprocessed finger vein sample by a key extraction matrix obtained by biological key training to obtain a fingerprint and finger vein fusion biological key; the method is characterized in that:
the method comprises the following specific steps:
step one, training a finger vein fusion biological key, which comprises the following specific steps:
firstly, a user respectively carries out sample collection on the same finger fingerprint and the finger vein for multiple times to obtain more than 3 fingerprint gray level images and finger vein gray level images, the fingerprint gray level images are uniformly scaled to 354 multiplied by 354 pixel size, the finger vein gray level images are scaled to 256 multiplied by 64 or 256 multiplied by 256 pixel size, and the fingerprint images obtained at this stage are marked as a first fingerprint image and a first finger vein image;
secondly, respectively carrying out equalization, convergence, smoothing, enhancement, binarization and refinement on the first fingerprint image and the first finger vein image obtained in the first step to obtain a preprocessed second fingerprint image and a preprocessed second finger vein image;
the first fingerprint image and the first finger vein image are respectively preprocessed to obtain preprocessed images, the preprocessing processes adopted by the two types of images are basically the same, the difference is only in the value difference of a plurality of parameters, and the preprocessing processes of the first fingerprint image and the first finger vein image are uniformly expressed as follows:
1) Respectively carrying out equalization processing on the first fingerprint image and the first finger vein image; a histogram equalization method, which is a common method in the field of image processing; the histogram equalization formula is shown in the following table:
wherein x represents a gray scale value between 0 and 255; f (x) represents that the gray value of the image is adjusted to f (x) at the point with the gray value of x; d max ＝255，A 0 Is the total number of pixels of the image, H i The number of points with the gray scale value i in the image is set;
2) Respectively carrying out convergence processing on the equalized first fingerprint image and the equalized first finger vein image obtained in the step 1); carrying out Gaussian filtering processing, namely image convergence processing, on the image by using a two-dimensional discrete Gaussian template operator; the method is a general method in the field of image processing; a plurality of sets of two-dimensional discrete Gaussian template operators G can be taken as examples
The computational formula is found in the following:
wherein: f (x, y) represents the gray value corresponding to the point with the coordinate (x, y) in the image after the equalization processing;
3) Respectively carrying out convergence processing on the first fingerprint image and the first finger vein image obtained in the step 2)Carrying out smoothing treatment on the image; smoothing the image by using a smoothing template operator; the method is a general method in the field of image processing; smooth template operator T may take multiple sets, taken as an example
The calculation formula is shown in formula (3):
wherein f' (x, y) represents the gray value corresponding to the point with the coordinate (x, y) in the fingerprint image and the finger vein image after smoothing;
4) Respectively carrying out image enhancement processing on the first fingerprint image and the first finger vein image which are obtained in the step 3) after the smoothing processing; carrying out image enhancement processing on the fingerprint image and the finger vein image after smoothing in the step 3) by using a Gabor function template; the method is a general method in the field of image processing; the Gabor function templates can take multiple sets, taken as an example:
determining the tangential direction (the direction perpendicular to the current direction) of a point with the coordinates (x, y) by using the value of VMAP [ x, y ], and taking 7 points adjacent to the point (x, y) in the tangential direction, wherein each point is represented by f (x, y); f (x, y) represents the gray value corresponding to the point with the coordinate (x, y) in the fingerprint image and the finger vein image; f (x, y) and Hw are subjected to convolution operation, and the operation result is assigned to f1 (x, y); f1 (x, y) is the gray value corresponding to the (x, y) point after the fingerprint image and the finger vein image are enhanced in the tangential direction after smoothing;
determining the normal direction of a point with coordinates (x, y) by using the value of VMAP [ x, y ] (the normal direction is obtained by clockwise rotating 90 degrees in the tangential direction), taking 7 f1 (x, y) points adjacent to the point (x, y) in the normal direction, performing convolution operation on the points and Vw, and assigning the operation result to f2 (x, y); f2 (x, y) is the gray value corresponding to the (x, y) point after the smoothed fingerprint image and the finger vein image are enhanced in the tangential line and normal line directions;
5) Respectively carrying out binarization processing on the enhanced first fingerprint image and the enhanced first finger vein image obtained in the step 4); the binarization processing is a general algorithm in the field of fingerprint and finger vein identification, and the binarization can be realized by directly calling functions such as im2bw () in matlab and the like;
6) Respectively thinning the first fingerprint image and the first finger vein image which are obtained in the step 5) and subjected to binarization processing; the field of image processing has a plurality of mature image thinning methods which can be used;
completing the steps 1) to 6), and obtaining a second fingerprint image and a second finger vein image after the pretreatment is completed;
thirdly, performing blind alignment operation on the second fingerprint image and the second finger vein image to obtain a third fingerprint image and a third finger vein image; although the sampling object is the same fingerprint, the sampling process has the problems of offset, rotation and the like, so that the images sampled for multiple times are not aligned, and the alignment operation is to perform the operations of offset, rotation and the like on the images sampled for multiple times, so that the positions of the acquired objects in the images are basically consistent; blind alignment means that the alignment process is completed only by using the self characteristic information of the current image, and is unrelated to the rest images without referring to the images;
the specific process of blind alignment of the fingerprint images comprises the following steps:
(1) Positioning a central point of the fingerprint image; the singular point locating process of the fingerprint image is a common method in the field of fingerprint identification, and is generally determined by calculating Poincare formulas, assuming that possible singular point coordinates are (i, j), calculating Poincare formulas (4), fife, (6):
where θ is the radian, O 'is the direction field (the direction field calculation is a general method in the art), O' (i + ε cos θ, j + ε sin θ) represents the direction field value after the change of the slight radian around the point (i, j),
the direction field difference value of the radian around the point (i, j) after slight change is represented, d delta represents the direction field change trend of the point (i, j) on the radian theta, poincare (i, j) integrates the direction field difference around the point (i, j), and then divides the integration by 2 pi, namely the average direction field difference around the point (i, j);
order to
Is determined as the center point of the fingerprint image; if a plurality of center points exist, taking the mass centers of the center points; if no central point exists, discarding the image, wherein the image cannot be used as a sample for training and testing;
(2) Translating the fingerprint image in the picture along the x-axis direction and the y-axis direction to enable the center point of the fingerprint image to be superposed with the center point of the picture;
(3) Taking the center point of the image, namely the center point of the fingerprint as the center of a circle, ml pixels as the radius to make a circle, wherein the specific value of ml is preferably that the edge of the circle can vertically cut fingerprint lines, the cutting point is marked as z, and generally, ml takes 15-30 pixels and is determined according to the actual condition; if no vertical cutting point exists, selecting the cutting point closest to the vertical cutting point and recording as z;
(4) Connecting the center point of the fingerprint with the point z, taking a ray, rotating the fingerprint image by taking the center point of the fingerprint as a circle center, and coinciding the ray made by the center point of the fingerprint and the point z with the y axis as a rotation result; the image rotation method is a general method in the image processing field;
so far, the blind alignment processing of the fingerprint images is finished; the fingerprint image blind alignment process is based on the following observations: a region is arranged near the center point of the fingerprint, the center point is used for making rays with the point of the region, and the direction of the rays is approximately parallel to the direction of the lines;
in the process of collecting the finger veins, as a bayonet is generally preset in the collecting equipment, fingers can prop against the top end of the bayonet, so that the positions of the collected finger veins in the length direction are basically fixed without alignment; in the width direction, as the fingers can move left and right in the acquisition process, the acquired images still have the possibility of being misaligned; accordingly, the blind alignment process of the finger vein images mainly considers the alignment in the width direction of the finger vein images;
the specific process is as follows:
(1) carrying out edge detection on the finger vein image, and extracting the finger vein edge image;
(2) the finger vein edge image is a binary image, has an upper edge, a lower edge and a left edge and a right edge, and is determined according to the placement condition of the image, and a linear regression operation is performed on a point set of the two edges to obtain two straight lines;
(3) placing the finger vein image in a Cartesian coordinate system, placing a certain vertex of an image rectangle at the origin of the coordinate system, wherein the length direction is parallel to the x axis, the width direction is parallel to the y axis, and the whole image is in the 1 st quadrant;
(4) calculating the center line of the two straight lines, wherein the calculation method comprises the following steps:
suppose that the coordinate equations of the two straight lines are y 1 ＝a1·x+b1，y 2 = a2 · x + b2, calculation
y 3 Is the central line of two straight lines, wherein a1, b1, a2 and b2 are real numbers, and y1 and y2 are linear equations in a Cartesian coordinate system;
(5) to calm the fingersThe pulse image is translated and rotated so that y 3 And a straight line
Overlapping; the specific method can firstly translate the finger vein image to enable y 3 ,y 4 Is coincident with the midpoint of the finger vein image, and then the finger vein image is rotated so that y 3 ,y 4 The two straight lines are completely overlapped; the image translation and rotation method is a common method in the field;
at this point, the blind alignment processing of the vein image is finished;
fourthly, extracting characteristic points from the third fingerprint image and the third finger vein image, recording characteristic point information, and sequentially arranging the characteristic point information into characteristic vectors; the characteristic points generally comprise end points, cross points and the like, pseudo end points and cross points are generally removed in the operation process, and coordinate values of the characteristic points are recorded from top to bottom and from left to right by using a two-dimensional Cartesian coordinate system to form characteristic vectors;
the endpoint extraction method comprises the following steps: scanning 8 points around the points in the thinned third fingerprint image and the third finger vein image, and if the sum of absolute values of differences of all two adjacent points of the 8 points is 2 multiplied by 255, the points are end points;
the method for extracting the cross points comprises the following steps: scanning 8 points around the points in the thinned third fingerprint image and the third finger vein image, and if the sum of absolute values of differences of all two adjacent points of the 8 points is 6 multiplied by 255, the points are cross points;
the end point and cross point extraction method is a general method in the field of fingerprint image processing;
fifthly, machine learning is carried out on the feature vector set obtained in the fourth step to obtain a fingerprint and finger vein fusion feature learning matrix W 1 I.e. a biometric key extraction matrix; fingerprint and finger vein fusion feature learning matrix W 1 The fingerprints and the finger vein features which are spliced in sequence are subjected to fusion calculation on a bit level, and the fused features are more stable than the original features;
collecting fingerprints and finger vein samples of a plurality of fingers of different users, repeatedly collecting a plurality of samples by the same finger, and obtaining a characteristic vector set after the four steps of operation; the set is divided into two categories, one category is that of the same finger of the user, and the other category is that of the non-user or the non-current finger, and is called as a positive and negative sample set;
with M = [ M = 1 ,M 2 ]Positive and negative sample sets, M, representing participation in training i ＝[x i1 ,x i2 ,...,x iL ]I ∈ {1,2} represents a set of class i samples, i =1 is a positive sample, i =2 is a negative sample; x is a radical of a fluorine atom ir ∈R d ,1≤i≤2,1≤r≤L，x ir Is a one-dimensional column vector, is a characteristic vector after fingerprint finger vein splicing, and is transposed to obtain a one-dimensional column vector x ir ，x ir Length d, R d Representing a d-dimensional real number domain, wherein L represents a sample characteristic vector obtained by sampling for L times in a sample set of the same finger, namely L column vectors;
now, according to the characteristics of the two types of samples, a fingerprint and finger vein fusion characteristic learning matrix W is trained 1 ∈R d×dz Obtaining the formula (7):
wherein
For a positive sample average of the training samples,
is the negative sample mean of the training sample; j is a cost function and reflects the learning matrix W of the fingerprint and finger vein fusion characteristics of the training sample 1 Calculating the distance difference between the projected image and the positive and negative sample set mean value by using Euclidean distance;
order:
solving matrix (H) 1 -H 2 ) Obtaining the fingerprint and finger vein fusion feature learning by the feature value and the feature vectorMatrix W 1 W 1 I.e. (H) 1 -H 2 ) w = λ w; w is a matrix (H) 1 -H 2 ) λ is a eigenvalue;
{w 1 ,w 2 ,...,w dz is a feature vector, corresponding to feature values { lambda } respectively 1 ,λ 2 ,...,λ dz In which λ is 1 ≥λ 2 ≥...≥λ dz ≧ 0, the eigenvectors with eigenvalues less than 0 are not included in the matrix W 1 The structure of (1); w is a group of 1 The fingerprint and finger vein fusion characteristic learning matrix is obtained;
the fingerprint and finger vein fusion feature learning part is completed to obtain a fingerprint and finger vein fusion feature learning matrix W 1 ；
Step two, fingerprint and finger vein fusion biological key extraction, which comprises the following specific steps:
step 1, collecting a fingerprint and a finger vein image by a user;
2, extracting a gray level image of the fingerprint and the finger vein image, wherein a color image can be used, and the color image is represented by a three-channel gray level image; carrying out equalization, convergence, smoothing, enhancement, binarization and refinement on the obtained fingerprint and finger vein images to obtain a refined image after preprocessing the fingerprint and finger vein images;
step 3, performing blind alignment operation on the fingerprint and the finger vein refined image to obtain an image after the fingerprint and the finger vein are blindly aligned;
step 4, extracting characteristic points from the image after the blind alignment of the fingerprint and the finger vein, wherein the characteristic points generally comprise end points, cross points and the like, and recording characteristic point information; the images after the blind alignment of the fingerprint and the finger vein are respectively placed in a two-dimensional Cartesian coordinate system, the coordinate values of the characteristic points are recorded from top to bottom and from left to right, and the coordinate values of each characteristic point are sequentially placed in a characteristic vector to form a fingerprint and finger vein fused characteristic vector x t ；
Step 5, learning a matrix W by using the fused features of the trained fingerprints and the finger veins of the first part 1 And after transposition, left-multiplying by the feature vector x obtained in the fourth step after fingerprint finger vein splicing t I.e. W 1 T ·x t D is obtained z Dimensional fusionPosterior feature vector x tz ；
Step 6, for x tz Each dimension component of (a) is subjected to a chessboard method operation, and the feature vector is further stabilized to
The chessboard method operation process is shown as formula (8):
Λ(x)＝k,(D+1)·k＜x tzi ≤(D+1)·k+D,(k＝0,1,…) (8)；
wherein D is the size of the grid of the chessboard method, a positive number is taken, a specific value can be selected by a user according to experience, the value of the lambda (x) is generally between 0 and 63, and x is tzi Is x tz Is quantized to an integer value; Λ (x) is x tzi The quantized value is the closest x in the checkerboard tzi Coordinate values of the grid of points and the origin of coordinates;
step 7, taking the vector of the calculation result of the step 6
The first n components of (1), n can be 16, 32, 64, etc., and is typically a power of 2, as the case may be
The number of effective characteristic components and the requirement of the biological secret key strength; splicing the n components back and forth to form a fingerprint and finger vein fusion biological key; if n is 64, each component takes a value of 0-64, 4-bit key calculation can be formed, and a result vector
The first n components of (a) may form a 256-bit key sequence;
thus, the fingerprint and finger vein fusion biological key is obtained.
2. A method for generating a biometric key by fusing a fingerprint and a finger vein at a bit level according to claim 1, wherein: the picture in the step (2) is an area with the same size as the fingerprint image, and is a rectangle with the size of 354 multiplied by 354 pixels, the central point of the area is positioned at the origin of a Cartesian coordinate system, and the length and the width are respectively parallel to the y axis and the x axis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811139551.0A CN109285204B (en) | 2018-09-28 | 2018-09-28 | Biological key generation method for fusing fingerprint and finger vein bit levels |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811139551.0A CN109285204B (en) | 2018-09-28 | 2018-09-28 | Biological key generation method for fusing fingerprint and finger vein bit levels |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109285204A CN109285204A (en) | 2019-01-29 |
| CN109285204B true CN109285204B (en) | 2022-11-25 |
Family
ID=65182419
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811139551.0A Active CN109285204B (en) | 2018-09-28 | 2018-09-28 | Biological key generation method for fusing fingerprint and finger vein bit levels |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109285204B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220130170A1 (en) * | 2019-02-14 | 2022-04-28 | Nec Corporation | Image processing device, fingerprint collation system, image processing method, and recording medium |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112906527B (en) * | 2021-02-05 | 2024-03-29 | 杭州电子科技大学 | Finger vein biological key generation method based on deep neural network coding |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103870810A (en) * | 2014-03-03 | 2014-06-18 | 杭州电子科技大学 | Method for generating human digital vein biometric key |
| CN104951774A (en) * | 2015-07-10 | 2015-09-30 | 浙江工业大学 | Palm vein feature extracting and matching method based on integration of two sub-spaces |
| WO2016011640A1 (en) * | 2014-07-24 | 2016-01-28 | 哈尔滨工业大学深圳研究生院 | Identification method based on handprint imaging |
-
2018
- 2018-09-28 CN CN201811139551.0A patent/CN109285204B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103870810A (en) * | 2014-03-03 | 2014-06-18 | 杭州电子科技大学 | Method for generating human digital vein biometric key |
| WO2016011640A1 (en) * | 2014-07-24 | 2016-01-28 | 哈尔滨工业大学深圳研究生院 | Identification method based on handprint imaging |
| CN104951774A (en) * | 2015-07-10 | 2015-09-30 | 浙江工业大学 | Palm vein feature extracting and matching method based on integration of two sub-spaces |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220130170A1 (en) * | 2019-02-14 | 2022-04-28 | Nec Corporation | Image processing device, fingerprint collation system, image processing method, and recording medium |
| US11893824B2 (en) * | 2019-02-14 | 2024-02-06 | Nec Corporation | Image processing device, fingerprint collation system, image processing method, and recording medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109285204A (en) | 2019-01-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3091479B1 (en) | Method and apparatus for fingerprint identification | |
| CN100557624C (en) | Face identification method based on the multicomponent and multiple characteristics fusion | |
| Feng et al. | Robust and efficient algorithms for separating latent overlapped fingerprints | |
| CN104951940B (en) | A kind of mobile payment verification method based on personal recognition | |
| Mathur et al. | Methodology for partial fingerprint enrollment and authentication on mobile devices | |
| Xu et al. | A fast minutiae-based fingerprint recognition system | |
| CN102332084B (en) | Identity identification method based on palm print and human face feature extraction | |
| Zhang et al. | Palmprint Recognition | |
| CN110443146A (en) | Auth method, device, equipment and readable medium based on bio-identification | |
| CN107679494B (en) | Fingerprint image matching method based on selective extension | |
| CN109285204B (en) | Biological key generation method for fusing fingerprint and finger vein bit levels | |
| CN109583279A (en) | A kind of fingerprint and refer to that vein combines recognizer | |
| CN107909031A (en) | A kind of scene of a crime fingerprint ridge leaves region frequency dynamic reconstruction method | |
| Gu et al. | Latent fingerprint registration via matching densely sampled points | |
| Raja et al. | Off-line signature verification based on fusion of grid and global features using neural networks | |
| CN107292273B (en) | Eight-neighborhood double Gabor palm print ROI matching method based on specific expansion | |
| Xu et al. | Complex spectral minutiae representation for fingerprint recognition | |
| CN109598235B (en) | Finger vein image authentication method and device | |
| US20080085039A1 (en) | Fingerprint matching method and apparatus using fingerprint minutiae and fingerprint binary images | |
| Vijayalakshmi et al. | Finger and palm print based multibiometric authentication system with GUI interface | |
| CN111753652B (en) | Three-dimensional face recognition method based on data enhancement | |
| CN105373781A (en) | Binary image processing method for identity authentication | |
| CN116342968B (en) | Dual-channel face recognition method and device | |
| CN110298275B (en) | Three-dimensional human ear identification method based on key points and local features | |
| Ribarić et al. | Personal recognition based on the Gabor features of colour palmprint images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |