TWI531985B - Palm biometric method - Google Patents

Description

Palm biometric method

The invention relates to a palm biometric identification method, in particular to a palm biometric identification method capable of effectively recognizing a palm image to be tested in any direction and at any angle.

Because of its low cost and high recognition rate, palm recognition technology has gradually become one of the mainstream technologies for identity recognition with biometrics.

"A biometric identification system based on eigenpalm and eigenfinger features" proposed by Ribaric, S. et al. in IEEE Transactions on Pattern Analysis and Machine Intelligence, wherein the prior art is controlled by a palm and its fingers. The majority of the sub-images are used to obtain the relevant feature values, and then the feature values are used for biometric identification. However, the prior art has specific requirements for the detected angles of the palm to be tested and the background brightness. Therefore, it is often easy to produce false positives in practical applications.

Wu, X., Zhang et al., "Palm line extraction and matching for personal authentication", IEEE Transactions on Systems, Man, and Cybernetics, 2006, wherein the prior art utilizes chaining coding (chaining coding). The technique is to obtain a plurality of palm lines for biometric identification. However, this prior art often makes it difficult to obtain such appropriate palm cutting lines because the sensitivity of the algorithm of the link encoding is too high.

In addition, Li, F., Leung et al., "Hierarchical identification of palmprint using line-based Hough transform" proposed in Proceedings of 18th Conference on Pattern Recognition, and Wu, J., Qiu et al. "A hierarchical palmprint identification method using hand geometry and grayscale distribution features" proposed in Proceedings of 18th Conference on Pattern Recognition, which utilizes palm geometry and its gray-scale distribution characteristics for performing Biometric identification, however, such prior art is unable to correct the angle of rotation of a palm image to be tested that is not vertically placed (i.e., a palm image having a rotational angle) to obtain the correct relative position of the palm.

Therefore, Ouyang, C.-S. et al., "An adaptive biometric system based on palm texture feature and LVQ neural network", Proceedings of 21th International Conference on Industrial, Engineering & Other Applications of Applied Intelligent Systems, 2006. According to a texture descriptor of a local fuzzy pattern (LFP), each sub-image of the palm to be tested is converted into a corresponding feature vector, and then utilized. The feature vector detects the rotation angle and position of the palm to be tested, although it can overcome the need for a specific angle and background brightness, the sensitivity of the algorithm is too high, has a rotation angle, etc., compared with the prior art described above. The problem, however, when the prior art scans the image of the palm to be tested, if the rotation angle of the palm to be tested is greater than 50 degrees, a misjudgment action may occur, so that a legitimate user cannot interpret the test for the palm identification system. A situation in which the palm image is unable to log in.

In addition, Ouyang, C.-S. et al., "An improved neural-network-based biometric system with rotation detection mechanism" proposed in International conference on machine learning and cybernetics (ICMLC) in 2010, although acceptable The image of the palm is scanned at an arbitrary rotation angle. However, in order to reduce the complexity of the recognition algorithm, the resolution of the palm image to be tested is reduced, and the corresponding subimages (Subimage) are normalized into a histogram, which will cause all The length and width of the finger image are missing. Therefore, although the prior art can increase the accuracy of identifying the position of the palm to be tested, a user with similar palm characteristics may occur because the missing portion of the corresponding sub-image is lost. The information causes the false positive rate of the palm recognition system to rise, so that the palm recognition system mistakes a legitimate user as an illegal user or an illegal user as a legitimate user.

Therefore, how to find a method that can improve the recognition of the image of the palm to be tested and effectively reduce the time required for the identification, so that the recognition accuracy and identification efficiency of the palm recognition system can be effectively improved, and it is worthy of continuous Research to improve goals.

Therefore, the object of the present invention is to provide a palm biometric method for receiving a palm image to be tested corresponding to a user's palm by a processor, comprising the steps of: assembling the processor to process the Measuring the palm image to obtain a set of boundary pixels; assembling the processor to detect the rotation angle of the palm to be tested; assembling the processor, sequentially searching each of the set of boundary pixels according to the direction of the palm to be tested a variation of the y value of a boundary pixel coordinate and the y value of the front and rear boundary pixel coordinates, wherein a plurality of feature points are set, wherein the feature points include a plurality of finger valley points, fingertip points, and target points; Obtaining a plurality of sub-images according to the feature points, comprising the following sub-steps: assembling the processor, according to a fingertip point and a corresponding two-finger valley point, or a fingertip point and a corresponding one of the target point and the one finger The valley point is calculated corresponding to the midpoint corresponding to each fingertip point, and then five interval points are respectively obtained according to the connection line between the midpoint and the corresponding fingertip point, and each interval point corresponds to the group boundary pixel get a crossover point, and the intersection point and the fingertip point, the midpoint, the finger valley point or the finger valley point and the target point are corresponding to the first sub-image; assembling the processor, according to a first set line segment formed by the two target points, and a square formed by one side of the first set line segment to obtain a specific point, and then obtained according to the specific point, the target points, and the corresponding valley points a second sub-image; assembling the processor to normalize the brightness of the sub-images; and assembling the processor to convert each sub-image into a corresponding feature vector; and assembling the processor, according to The feature vectors obtain and store a bio-representative model corresponding to the user.

The above and other technical contents, features and advantages of the present invention will be apparent from the following detailed description of the preferred embodiments.

According to Ouyang, C.-S. et al., "An adaptive biometric system based on palm texture feature and LVQ neural network" proposed by Proceedings of 21th International Conference on Industrial, Engineering & Other Applications of Applied Intelligent Systems, 2006. In the technology, a palm biometric method includes the following procedures: a registration procedure and an identification and verification.

The registration procedure includes receiving a palm feature vector of a set of legitimate users, and performing a learning vector quantization (LVQ2) algorithm operation to obtain a bio-representative model corresponding to the legal user ( Representative prototype), and store the bio-representative models in a Registry user database, where each bio-representative model has six feature vectors.

The verification program includes receiving a palm eigenvector of a group of unknown users, and comparing the palm eigenvectors with six eigenvectors of each biometric model stored in the legitimate user database, if one exists When the six feature vectors of the biological representative model are equal to the palm feature vector of the unknown user, the unknown user will be regarded as a legitimate user and accept the login request. If not, the unknown user will be Rejected as a illegitimate user and denied their login request.

Therefore, the features of the present invention focus on how to effectively obtain a palm eigenvector of a group of legitimate users, convert it into a corresponding bio-representative model and store it, or obtain a set of palm eigenvectors of an unknown user. The six eigenvectors of each biometric model in the legal user database are compared to determine whether the unknown user is a legitimate user.

Referring to FIG. 1 , a preferred embodiment of the palm biometric identification method 9 of the present invention is adapted to receive, by a processor, a palm image to be tested corresponding to a user's palm, and the method includes the following steps: Step 91 is to assemble the processing. The processor is configured to process the image of the palm to be tested; step 92 is to assemble the processor to detect the rotation angle of the palm to be tested; and step 93 is to assemble the processor to search for all feature points of the palm to be tested; And step 94 is to assemble the processor to obtain a feature vector of the palm to be tested according to the feature points.

Each step 91-94 is described as follows: Referring to FIG. 2, step 91 is a step of processing the image of the palm to be tested, and has the following sub-steps: sub-step 911 is to assemble the processor to set a grayscale pixel threshold ( Threshold of gray level)T, and according to the grayscale pixel threshold, the palm image to be tested is converted into a binary image, wherein the grayscale pixel threshold T is set as follows ( F .1) shown:

M and N represent the resolution of the palm image to be tested as M×N, and g ( x , y ) represents the pixel value (pixel value) of the coordinates of ( x , y ) in the image of the palm to be tested.

In addition, the conversion manner of the two-bit image is: when the processor determines that the pixel value g ( x , y ) of the pixel whose coordinates are ( x , y ) in the image of the palm to be tested is greater than the threshold value T of the gray-scale pixel When the processor is configured to set the pixel value g ( x , y ) to 255, and when the pixel value g ( x , y ) is less than the gray-scale pixel threshold T, the processor is configured to The pixel value g ( x , y ) is set to 0; the sub-step 912 is to assemble the processor, and to eliminate the noise point in the binary image according to a 3×3 median filter; Step 913 is to assemble the processor, and use a Laplacian filter to complete edge detection in the binary image, thereby obtaining a set of boundary pixels, wherein the group Each boundary pixel in the boundary pixel is one of the pixels on the palm contour of the image of the palm to be tested; and sub-step 914 is to assemble the processor to sequentially mark the set of boundary pixels in the sub-step 914. The order is: (M-1, N-1), (M-1, N-2), ..., (M-1, 0), (M-2, N-1), (M- 2, N-2), ..., (0, 0).

Referring to FIG. 3 and FIG. 4 together, step 92 is a step of detecting a rotation angle of the palm, and has the following sub-steps: sub-step 921 is to assemble the processor, set a horizontal line segment in which a plurality of strips are interlaced with the image of the palm to be tested, and determine when When a value on a horizontal line segment is equal to a pixel coordinate value in the set of boundary pixels, that is, the horizontal line segment has an intersection with the palm contour in the image of the palm to be tested; and sub-step 922 is to assemble the processor. Selecting a horizontal line segment having eight interlaced points P1~P8 with the palm outline as a first reference line segment; sub-step 923 is assembling the processor, selecting the center point PM1 of the interlaced points P2 and P3, and Selecting a corresponding reference point P _ref from the set of boundary pixels according to the center point PM1; sub-step 924 is to assemble the processor, and determining the direction of the palm to be tested according to the reference point P _ref coordinate and the interlaced point P2 The y coordinate of the reference point P _ref coordinate is larger than the y coordinate of the interlaced point P2, which means that the direction of the palm to be tested is the direction of the finger tip toward the -y direction, and vice versa, when the y coordinate of the reference point P _ref coordinate is smaller than the interlace Point P2 a coordinate indicating that the direction of the palm to be tested is the direction of the tip of the finger toward the +y direction; the sub-step 925 is to assemble the processor, and a first marker point is set according to the interlaced points P3, P4, wherein the direction of the palm to be tested is When the direction of the tip of the finger is toward the +y direction, the first marker point is set by calculating the point PM2 among the interlaced points P3 and P4, and then calculating the boundary pixel of the group, the x coordinate is between the interlaced points P3 and P4. The distance between the y coordinate and the y coordinate of the y coordinate of the midpoint PM2 and the midpoint PM2, wherein the pixel having the largest distance is the first marker point, similarly, if the direction of the palm to be tested is a finger When the tip direction is in the -y direction, the distance between each pixel of the set of boundary pixels, the x coordinate is between the interlaced points P3, P4, and the y coordinate is smaller than the midpoint PM2, and the midpoint PM2 is calculated. The pixel having the largest distance is the first marker point; similarly, the processor is configured to set a second marker point according to the interlace points P5 and P6 and the point PM3 among the interlaced points P5 and P6; Sub-step 926 is to assemble the processor to calculate the first mark separately The slope of the midpoint PM2 is a first slope, the slope of the second marker point and the midpoint PM3 is a second slope, and the average of the first slope and the second slope is calculated as an average slope; Step 927 is to assemble the processor, and according to the average slope, a corresponding rotation angle can be obtained. For example, when the average slope is 1, according to the Bishop's theorem, the corresponding rotation angle is 45 degrees, and the other So on and so forth.

Referring to FIG. 5 and FIG. 6, step 93 is a step of searching for palm feature points, and has the following sub-steps: sub-step 931 is to assemble the processor, and sequentially search for each boundary of the set of boundary pixels according to the direction of the palm to be tested. y y values change in the case of the pixel values of boundary pixels of the front and rear, the palm when the test direction towards the finger tip + y direction, a certain boundary pixels S _n y values are smaller than the previous boundary pixels S _n-1 and after a boundary pixel S _{n +} y value of _1, the object boundary pixels S _n is defined as a finger valley points (Valley-point), while the y value if a certain boundary pixels S _m are each greater than the previous one boundary pixel S The y value of _m-1 and the latter boundary pixel S _m+1 , the target boundary pixel S _m is defined as a fingertip-point, so that five fingertip points T1 to T5 and four can be finally obtained. Pointing to the valley point B1~B4, wherein the second fingertip point T2 is the first marking point, and the third fingertip point T3 is the second marking point; otherwise, when the direction of the palm to be tested is a finger -y direction towards the tip, y boundary values of a target pixel is smaller than S _n are respectively a front boundary pixels S _n-1 as a boundary and the rear The y value of the prime S _n+1 , the target boundary pixel S _n is defined as a fingertip point, and if the y value of a target boundary pixel S _m is greater than the previous boundary pixel S _m-1 and the subsequent boundary pixel S respectively _{m +} y value of _1, the target pixel boundary is defined as S _m a valley means; and a sub-step 932 is set with the processor, the first valley point B1 and the distance of the fingertip based on the first point T1 Obtaining a boundary pixel having an equal distance in the set of boundary pixels, and defining it as a first target point K1, similarly, assembling the processor according to the fourth fingertip point T4 and the third finger valley Point B3, obtaining a boundary pixel having an equal distance in the set of boundary pixels, and defining it as a second target point K2, and according to the fifth fingertip point T5 and the fourth finger point B4, The group boundary pixel has one of the equal distance boundary pixels and is defined as a third target point K3.

Therefore, after this step is completed, five fingertip points T1~T5, four finger points B1~B4, and three target points K1~K3 are obtained, and a total of twelve palm feature points are obtained.

Referring to FIG. 7, step 94 is a step of obtaining the palm feature vector to be tested, and has the following sub-steps: step 941 is to assemble the processor to obtain six sub images, including five finger images and one a palm image, and the processor obtains a first finger image according to the first target point K1, the first finger point B1, and the first finger point T1, according to the first and second finger points B1. B2 and the second fingertip point T2 obtain a second finger image, and according to the second and third finger points B2 and B3 and the third fingertip point T3, a third finger image is obtained according to the second target. a fourth finger image is obtained from the point K2, the third finger point B3, and the fourth finger point T4, and is obtained according to the third target point K3, the fourth finger point B4, and the fifth finger point point T5. a fifth finger image; referring to FIG. 8, the first finger sub-image is described as being set as follows: the processor calculates the first target point K1 and the point m1 of the first finger point B1, and according to a line segment at which the midpoint m1 and the first fingertip point T1 are both end points, respectively, at a length of one sixth of the line segment , at the length of the two-sixth line, the length of the three-sixth line, the length of the six-fifth line, and the length of the five-fifth line, set a spacing point m2, m3, m4, m5, m6, and then The interval points m2~m6 respectively extend over the contour of the first finger to a set of crossing points (F1, F2), (F3, F4), (F5, F6), (F7, F8), (F9, F10), therefore, sequentially along the first target point K1, the intersection points F1, F3, F5, F7, F9, the first fingertip point T1, the intersection points F10, F8, F6, F4, F2, the first finger point B1 and the midpoint m1 obtain a closed first finger sub-image.

The rest of the finger images are obtained in the same manner as the first finger image, and will not be described here.

In addition, referring to FIG. 9, the palm image is set as follows: the processor obtains a first set line segment according to the first target point K1 and the second target point K2, and then uses the first set line segment. Forming a square for one side to obtain a specific point S1 connected to the first target point, and finally, assembling the processor to sequentially follow the specific point S1, the first target point K1, the first The valley point B1, the third index point B3, the second target point K2, and the third target point K3 are obtained to obtain a closed palm image.

Therefore, referring to FIG. 10, the corresponding five finger images and one palm image can be obtained.

Referring to FIG. 7, step 942 is to assemble the processor, normalize the brightness of the sub-images according to a histogram equalization method, so that the brightness of each sub-image is relatively uniform; and step 943 is a group. The processor is configured to convert a sub-image of each palm to be tested into a corresponding feature vector according to a texture descriptor of a local fuzzy pattern (LFP), and a conversion method thereof is described. As follows: Suppose each pixel point P ( x , y ) of a sub-image, and its corresponding luminance value is I ( x , y ), and the luminance value set of P ₁ reference blocks in its neighborhood N ₁ is { I ( x _i , y _i )|1 i P ₁ ). First, calculate the following vectors:

And μ _i (1)=1/(1+exp{-α( I ( x _i , y _i )- I ( x , y ))}), μ _i (0)=1-μ _i (1), Where μ _i (1) and μ _i (0) are respectively a fuzzy attribution function to represent the difference in luminance between the reference pixel P ( x _i , y _i ) and the pixel P ( x , y ) in a neighborhood The value is encoded as a degree value of 1 and 0. Next, calculate the following vector H ( x , y ):

H ( x , y )=(ω ₁ ,ω ₂ ,...,ω _Q ).....( F .2)

Where ω _j = μ ₁ ( b ₁ ) × μ ₂ ( b ₂ ) × ... × , Q = -1, b ₁ b ₂ ... P j for the first group the binary digits _{1 yuan,} the total Q = - 1 set of possible binary combinations. Therefore, ω _j represents the degree of value encoded into b ₁ b ₂ ... b _P . Then normalize it to obtain the normalized vector H' ( x , y ) = ( ). Finally, by adding and normalizing H' ( x , y ) of all the pixels in the subgraph, the feature vector corresponding to the sub image can be obtained.

In summary, the present invention can effectively recognize the image of the palm to be tested input at any position and at any angle, and by accurately setting the sub-images, the more accurate feature vectors can be correspondingly obtained to reduce the palm recognition. The false positive rate of the system makes it possible to achieve the object of the present invention.

The above is only the preferred embodiment of the present invention, and the scope of the invention is not limited thereto, that is, the simple equivalent changes and modifications made by the scope of the invention and the description of the invention are All remain within the scope of the invention patent.

9. . . Palm biometric method

91~94. . . step

911~914. . . Substep

921~927. . . Substep

931~932. . . Substep

941~943. . . Substep

Figure 1 is a flow chart of a preferred embodiment of the present invention;

2 is a flow chart of the substep of processing the image of the palm to be tested;

Figure 3 is a flow chart of the substep of detecting the rotation angle of the palm;

4 is a schematic diagram of an example of detecting a rotation angle of a palm;

Figure 5 is a flow chart showing the substeps of the search for palm feature points;

Figure 6 is a schematic view of the feature points of the palm to be tested;

7 is a flow chart of the substep of obtaining the palm feature vector to be tested;

Figure 8 is a schematic diagram of setting a finger image;

Figure 9 is a schematic diagram of setting a palm image; and

FIG. 10 is a schematic diagram of the sub-image obtained by the preferred embodiment.

9. . . Palm biometric method

91~94. . . step

Claims (4)

A palm biometric identification method is adapted to receive a palm image to be tested corresponding to a user's palm by a processor, comprising the steps of: assembling the processor to process the image of the palm to be tested to obtain a set of boundary pixels And assembling the processor to detect the rotation angle of the palm to be tested, comprising the following substeps: assembling the processor, selecting, according to a center point of two interlaced points in a first reference line segment, by the group of boundary pixels a corresponding reference point; assembling the processor, determining the direction of the palm to be tested according to coordinates of the reference point coordinates and the interlaced points; and assembling the processor, respectively setting one according to the adjacent two interlaced points a first marker point and a second marker point, and calculating the corresponding midpoints, and then calculating a first slope and a second slope according to the midpoints and the first and second marker points, And obtaining an average slope from the average of the first and second slopes; and assembling the processor to obtain a corresponding rotation angle according to the average slope; assembling the processor according to the direction of the palm to be tested, in order Search a plurality of feature points are set, wherein the feature points include a plurality of finger points, fingertip points, and target points, wherein a y value of each boundary pixel coordinate of the set of boundary pixels and a y value of the front and rear boundary pixel coordinates are changed; The processor is assembled to obtain a plurality of sub-images according to the feature points For example, the method includes the following substeps: assembling the processor, calculating a corresponding finger according to a fingertip point and a corresponding two finger valley point, or a fingertip point and a corresponding one of the target point and the one finger valley point The cusp corresponds to the midpoint, and then five gap points are respectively obtained according to the connection between the midpoint and the corresponding fingertip point, and each interval point correspondingly obtains two crossing points in the set of boundary pixels, and the same a first sub-image of the intersection point and the fingertip point, the midpoint, the finger valley point or the finger valley point and the target point; the processor is assembled, and one of the two target points is formed a first set line segment, and a square formed by one side of the first set line segment to obtain a specific point, and then obtaining a second sub-image according to the specific point, the target points, and the corresponding valley points; The processor is configured to normalize the brightness of the sub-images; and the processor is configured to convert each sub-image into a corresponding feature vector; and the processor is assembled, according to the special micro-vectors And storing the bio representative model corresponding to the user. According to the palm biometric identification method of claim 1, wherein in the step of detecting the rotation angle of the palm to be tested, when the y coordinate of the reference point coordinate is larger than the y coordinate of the interlaced point, determining the The direction of the palm to be tested is the tip of the finger toward the -y direction. When the y coordinate of the coordinate of the reference point is smaller than the y coordinate of the interlaced point, the direction of the palm to be tested is The tip of the finger is oriented in the +y direction. According to the palm biometric identification method of claim 1, wherein in the step of detecting the rotation angle of the palm to be tested, the first reference line segment is a horizontal line having eight interlaced points with the image of the palm to be tested. segment. According to the palm biometric identification method of claim 1, wherein the step of searching for all the feature points of the palm to be tested further comprises: assembling the processor according to a fingertip point and a corresponding finger point point. Distance, setting a boundary pixel having an equal distance in the set of boundary pixels as a target point.