CN105938549A - Palm print ROI segmentation method in palm print identification - Google Patents

Abstract

The invention discloses a palm print ROI segmentation method in palm print identification, and belongs to the field of palm print identification, for solving the problems of difficult determination of positioning points in a square-based positioning segmentation method of palm print segmentation and quite large similar image ROI extraction deviation in a conventional palm print identification process. The method comprises the following steps: S1, selecting a fitting line; and S2, performing image rectification and palm print ROI segmentation. The method has the following advantage: an image segmentation algorithm reduces the influences of image rotation and translation during image acquisition.

Description

Palmmprint ROI dividing method in personal recognition

Technical field

The invention belongs to personal recognition field, relate to the palmmprint ROI dividing method in a kind of personal recognition.

Background technology

Along with development and the raising of scientific and technological level of society, the safety consciousness of the people constantly strengthens, and the safety of information is subject to Concern increasingly, therefore in actual life, everyone is often in the face of the discriminating problem of identity.Traditional authentication warp Frequently with password, password, certificate etc., there is the biggest drawback in these traditional discrimination methods.Biometrics identification technology is because of it The advantage that inherently safe grade is high, the most slowly replace traditional identity identifying method, through frequently with fingerprint, face, iris, The characteristics of human body such as gait, person's handwriting, hand, palmmprint.At present, single biological characteristic has the limitation that it is intrinsic, does not also have a kind of single Only biometrics identification technology can meet the demand of reality.Multi-modal biological characteristic identification technology is melted by multi-biological characteristic The method closed, improves the accuracy rate of identification and expands range of application, with the demand of satisfied reality.Due to hand images collection side Just, user's acceptance is high, comprise contain much information, recognition accuracy more high, be widely used at present.

Personal recognition generally comprises several major parts such as palmmprint extraction, palmprint information analysis, during wherein palmmprint extracts, and meeting Relate to palmmprint segmentation step, prior art based in foursquare locating segmentation method, anchor point is difficult to determine and similar It is bigger that image ROI extracts drift rate.

Summary of the invention

During solving existing personal recognition identification, palmmprint segmentation had based on foursquare locating segmentation In method, anchor point is difficult to determine and extracts the bigger problem of drift rate with similar image ROI, and the present invention proposes a kind of personal recognition In palmmprint ROI dividing method, with easier realization for the determination of anchor point in foursquare locating segmentation method, and can To reduce image ROI extraction drift rate, to achieve these goals, the technical scheme is that

Palmmprint ROI dividing method in a kind of personal recognition, comprises the steps:

S1. fitting a straight line is chosen；

S2. the ROI segmentation of image flame detection and palmmprint.

Further, choose the reference direction that a stable straight line is split as ROI in the picture, for profile diagram The marginal point of one quadrant uses least square fitting to go out straight line.

Further, determine the central point of ROI, with valley point M1 as fixing point, find on a M2 is expert at and make straight line M1M2 Become the some M2 ' of fixed angle, the midpoint O of line taking section M1M2 ' with fitting a straight line L, do the perpendicular bisector of straight line M1M2 ', and The right side area of perpendicular bisector finds the some O1 of regular length R, then some O1 is just in the central area of palm, finally with an O1 For the central point of ROI, the square area of intercepting 128 × 128 is as the ROI of image, describedDescribed just Square two of which limit is parallel to fitting a straight line L.

Further, the step of fitting a straight line is:

If the equation expression formula of straight line is:

Y=kx+b (1)

Measured value according to volar edge profile obtains straight line intercept b on the y axis and straight slope k, (x_i,y_i) it is hands The coordinate of the measured value of palm edge contour, b₀、k₀For the approximation of b, k, make:

B=b₀+δb

K=k₀+δk

Wherein, δ b and δ k is the deviation of slope and intercept；

Using y as dependent variable, with x as independent variable, error equation is:

The matrix expression of error equation is:

A δ X=L+V

Wherein

By method of least square criterion

V^TV=min

I.e.

Its least square solution is:

The value drawing k, b with this, brings formula 1 into and i.e. obtains fit equation and fitting a straight line.

Beneficial effect: algorithm solves and is difficult to determine and similar image based on anchor point in foursquare locating segmentation method ROI extracts the bigger problem of drift rate, image rotation and the impact of translation when this image segmentation algorithm also reduces image acquisition. Algorithm is to solve challenge by straightforward procedure, and in the case of reaching same effect compared with existing additive method, algorithm is not Only saving the time and be more easily implemented, and the ROI drift rate extracted is less, algorithm is reliable, has more practicality.

Accompanying drawing explanation

Fig. 1 is the hand shape image that processes of the present invention and hand-type characteristic point position schematic diagram；

Fig. 2 is disk algorithm schematic diagram of the present invention；

Fig. 3 is hand local block schematic diagram of the present invention；

Fig. 4 is palmprint image and ROI segmentation figure.

Detailed description of the invention

Embodiment 1: in personal recognition, most important step is exactly the segmentation of palmmprint region of interest (ROI), for original calculation The defect of method, proposes a kind of ROI dividing method based on specific part fitting a straight line.The contour line of palm can open along with finger Degree change, and the contour line of the marginal area of palm little finger side will not change with the change of finger stretching degree. According to this feature, least square fitting is used to go out straight line L for the specific marginal point of palm profile.With straight line L it is With in Fig. 4 (a) two, benchmark, refers to that valley point M1, M2, as reference point, are two straight line ab and straight line cd being parallel to straight line L respectively；With The straight line OO1 being parallel to straight line L is at the midpoint of some M1, M2, is perpendicular to the straight line of L by a M1, and this straight line is handed in straight line cd Point is M2 ', is O1 in straight line OO1 intersection point, on the basis of an O, determines that a certain length intercepts on straight line OO1, determines an O1.With Centered by some O1, determine intercepted length, in the direction parallel and perpendicular to straight line L, image is separated respectively, it is thus achieved that the palm Stricture of vagina ROI, as shown in Fig. 4 (a).The present embodiment describes the palmmprint ROI dividing method during a kind of personal recognition, including as follows Step:

1) fitting a straight line is chosen

Choose the reference direction that a stable straight line is split as ROI the most in the picture.By the analysis to image Find, when gathering image, although there is the randomness that finger opens, but the profile in the back edge region of palm little finger side Line varies less, and according to this feature, the marginal point for profile diagram first quartile uses least square fitting to go out one directly Line.

If the equation expression formula of straight line is:

Y=kx+b (1)

Optimal b (straight line intercept on the y axis) and k (straight slope) obtained by measured value according to volar edge profile. (x_i,y_i) it is the coordinate of the measured value of volar edge profile, b₀、k₀Approximation for b, k.Order:

B=b₀+δb

K=k₀+δk

Using y as dependent variable, with x as independent variable, error equation is:

Wherein, δ b and δ k is the deviation of slope and intercept；

The matrix expression of error equation is:

A δ X=L+V

Wherein

By method of least square criterion (min represents minima)

V^TV=min

I.e.

Its least square solution is:

Thus draw the value of a, b, and to bring formula 1 into and i.e. obtain fit equation, matching as required by Fig. 4 (a) cathetus L is exactly is straight Line.

2) the ROI segmentation of image rectification and palmmprint

After palmprint image is carried out above process, start to determine the central point of ROI.For reducing same person image center Offset problem adopt with the following method.As shown in Fig. 4 (a), with valley point M1 as fixing point, find on a M2 is expert at and make straight line M1M2 becomes the some M2 ' of fixed angle (90 degree taken in experiment) with fitting a straight line L.The midpoint O of line taking section M1M2 ', does straight line The perpendicular bisector of M1M2 ', and find regular length R (wherein in the right side area of perpendicular bisector) Point O1, then some O1 is just in the central area of palm, and the finally central point with an O1 as ROI intercepts the square of 128 × 128 (two of which limit is parallel to fitting a straight line L) region is as the ROI of image.Fig. 4 (b) is the innovatory algorithm segmentation to particular image Experiment simulation figure.

The present embodiment proposes a kind of new locating segmentation algorithm for the deficiency in existing method, algorithm solve based on In foursquare locating segmentation method, anchor point is difficult to determine and extracts the bigger problem of drift rate, this image with similar image ROI Image rotation and the impact of translation when partitioning algorithm also reduces image acquisition.Algorithm is to solve challenge by straightforward procedure, In the case of reaching same effect compared with existing additive method, algorithm is not only saved the time but also is more easily implemented, and carries The ROI drift rate taken is less, and algorithm is reliable, has more practicality.

Embodiment 2: present embodiment discloses a kind of multimodal Biometrics method based on hand and palmmprint, wherein, hands Shape identification includes several major parts such as hand contours extract, positioning feature point, characteristic quantity analysis.And personal recognition generally comprises Several major parts such as palmmprint extraction, palmprint information analysis, during wherein palmmprint extracts, can relate to the step of palmmprint segmentation.For The part of palmmprint, such as the record of technical scheme in embodiment 1, and for the record of hand part, refers to following proposal.This Outward, the record of described hand part, can be higher level's step or subordinate's step of the record of palmmprint part, divide as palmmprint ROI A part for segmentation method.

Hand shape image is done gray proces, carries out grey level enhancement；Determine segmentation threshold, image is carried out binaryzation；Pass through Frontier tracing, extracts hand profile as shown in Figure 1.By the analysis to Fig. 2, on contour line, certain point is as the center of circle, is half with R Footpath, in circle, the existing target area pixel that belongs to also has and belongs to background area pixels point.Can be seen that when disk moves on straight line Time dynamic, in disk, the point of some target area and background area is above the center of circle, and some is in the lower section in the center of circle.And work as disk When forwarding the flex point of convex domain to, disk region of interest within the most all in the lower section of centre point, when disk forwards to down convex During the flex point in region, in disk background area the most all above centre point.Disk is proposed based on above theory Extreme value algorithm, hand contour line is internal is target area, and outside is background area, permissible by analyzing hand profile diagram (Fig. 1) Find out, it is assumed that disc centre point T at a certain Fingers peak, then the point in the neighborhood around a T is all in its lower section Or same a line, for referring to that paddy also has similar feature, unique unlike point in neighborhood referring to above valley point or Same a line, and only refer to peak and refer to that paddy characteristic point has this feature, so that it is determined that Fingers peak dot and the position of finger valley point.

In Fig. 3 (a), determining that middle finger refers to peak dot place smaller area, middle finger refers to peak dot T2 to utilize disk extremum method to determine, With T2 column, hand shape image being divided into two parts, Fig. 3 (b) is nameless little finger region subgraph, and Fig. 3 (c) is food Refer to region subgraph.In Fig. 3 (b), determine finger region, valley point between little finger and the third finger, utilize disk extremum method true Fixed this refers to valley point T7.For Fig. 3 (c), determine partitioning parameters, be slit into forefinger and middle interphalangeal refer to region, valley point subgraph 3 (d) and Forefinger refers to peak dot region subgraph 3 (e).Disk extremum method is utilized to determine forefinger respectively in the less region of Fig. 3 (d) Fig. 3 (e) Refer to that valley point T5 and forefinger refer to peak dot T1 with middle interphalangeal.Further determine that partitioning parameters, Fig. 3 (b) is divided into middle finger with nameless Between refer to region, valley point subgraph 3 (f), the third finger refer between peak dot region subgraph 3 (g) and little finger refer to peak dot region subgraph 3 (h).? Disk extremum method is utilized to determine finger valley point T6 between middle finger and the third finger, at Fig. 3 (g) and Fig. 3 (h) in region less in Fig. 3 (f) In utilize the disk extremum method third finger respectively to refer to that peak dot T3 and little finger refer to peak dot T4 in less region.

Hand shape image does gray processing process, draw the rectangular histogram of gray level image, find out pixel grey scale and concentrate scope, carry out Grey level enhancement, makes image become apparent from.Use local threshold binaryzation, the circle using radius to be 1 again the image after binaryzation Dish carries out corroding dilation operation, rejects zonule, can carry out feature location afterwards,

In the step of feature location, the present embodiment proposes the side of characteristic point stationary positioned order in a kind of hand identification Method, makes technical term in this method and being defined below: subgraph b is nameless little finger region subgraph, and subgraph c is forefinger Region subgraph, subgraph e is that forefinger refers to peak dot region subgraph, and subgraph f is to refer to region, valley point between middle finger and the third finger Figure, subgraph g is that the third finger refers to peak dot region subgraph, and subgraph h is to refer to peak dot region subgraph between little finger；

Described method comprises the steps:

S1. 7 empty arrays S are created_i[] is used for depositing the finger peak belonging to same finger meeting condition or the spy referring to paddy Levy a little, wherein: i=1 ..., 7；

S2. artwork a is carried out the intersection point first of scanning from top to bottom, from left to right, search sweep line and finger, with this On the basis of Dian, contour line point all below the center of circle is stored in array S to utilize disk extremum method to determine₁In, array S₁Intermediate point It is exactly that middle finger refers to peak dot T2；

S3. refer to that artwork is divided into subgraph b and subgraph c by peak dot T2 according to middle finger, to subgraph from bottom to top, swept by left-to-right Retouch, when there is multiple intersection point first in scan line and contour line, with this row except with the intersection point of left side edge contour line in addition to its Point on the basis of its intersection point, utilizes disk extremum method to determine contour line point poke group S all below the center of circle₂In, array S₂In Between point be exactly little finger and nameless finger valley point T7；

S4. calculateWherein x₂、x₇For the abscissa of T2, T7, to subgraph c with n₃For left margin it is It is the region of subgraph e, subgraph e is carried out the intersection point first of scanning from top to bottom, from left to right, search sweep line and finger, On the basis of this puts, contour line point all below the center of circle is stored in array S to utilize disk extremum method to determine₃In, array S₃In Between point be exactly that middle finger refers to peak dot T2；

S5. calculatex₁For the abscissa of T1, to subgraph d, row is by y₇Upwards, row are by x₂To n₄'s Region is scanned, search sweep line and the intersection point first of finger, on the basis of this puts, utilizes disk extremum method to determine contour line All the point below the center of circle is stored in array S₄In, array S₄Intermediate point be exactly the finger valley point T5, wherein y of forefinger and middle interphalangeal₇ It it is the vertical coordinate of a T7；

S6. calculatex₅For the abscissa of T5, to subgraph f, row is by y₇Upwards, row are by n₅To x₂'s Region is scanned, search sweep line and the intersection point first of finger, on the basis of this puts, utilizes disk extremum method to determine contour line All the point below the center of circle is stored in array S₅In, array S₅Intermediate point be exactly the finger valley point T6 between middle finger and the third finger；

S7. calculateTo subgraph b with n₆It is i.e. the region of subgraph g for right margin, subgraph g is carried out Scanning from top to bottom, from left to right, search sweep line and the intersection point first of finger, on the basis of this puts, utilize disk extreme value Method determines that contour line point all below the center of circle is stored in array S₆In, array S₆Intermediate point be exactly that the third finger refers to peak dot T3；

S8. according to fixed calculating y_max=MAX (y₁,y₃), y_min=MIN (y₁,y₃), a₃=| y₂-y_min|, antithetical phrase Figure h, row is by (y_max+a₃) downwards, row are with n₆Region for right margin is scanned, and record intersection point is more than the line number of 2 first, will Meet | n_i-n_i+1| the intersection point of >=2 is stored in array S₇In, array S₇Intermediate point be exactly characteristic point T4 that little finger refers to peak dot.

Wherein:

Subgraph b is nameless little finger region subgraph, and subgraph c is forefinger region subgraph, and subgraph e is that forefinger refers to Peak dot region subgraph, subgraph f is to refer to region, valley point subgraph between middle finger and the third finger, and subgraph g is that the third finger refers to peak dot region Subgraph, subgraph h is to refer to peak dot region subgraph between little finger；

n₃Forefinger refers to peak dot dividing sub-picture parameter, n₄Forefinger and middle interphalangeal refer to valley point dividing sub-picture parameter, n₅Middle finger and unknown Valley point dividing sub-picture parameter, n is referred between finger₆The third finger refers to peak dot dividing sub-picture parameter.

y₁, y₂, y₃It is respectively characteristic point T1, the vertical coordinate of T2 and T3, y_maxFor y₁And y₃Maximum, y_minFor y₁And y₃'s Minima.

Palmmprint ROI dividing method in above-mentioned personal recognition, because have employed specific region fitting a straight line and fixed character Point location technology, it is possible to fast and effeciently extract palmmprint ROI.Overcome the deficiency of original algorithm, figure when reducing image acquisition As rotating and the impact of translation.In terms of computational efficiency with accuracy rate, there is greater advantage compared with original algorithm, calculate the time Being greatly shortened, and be more easily implemented, the realization for identity authorization system based on palmmprint provides theory and experimental basis.Should Algorithm not only accuracy rate is high, speed is fast, algorithm is simple and solve that traditional method surface sweeping scope is big, disk threshold value and radius difficulty With the difficult problem determined, feature location effect significantly improves, and algorithm also reduces the requirement to image acquisition, improves user simultaneously Comfortableness, gathered person's finger stretching degree is not had rigors, the user of defective to finger (bending, excalation) It also is adapted for this algorithm.

Additionally, the method for characteristic point stationary positioned order in the hand identification that relates to of such scheme, have employed hand shape image Partition, utilizes disk extreme value algorithm, can extract hand-type characteristic point fast and accurately, and this algorithm not only accuracy rate is high, fast Degree is fast, algorithm simple and solve that traditional method surface sweeping scope is big, disk threshold value and radius are difficult to the difficult problem that determines, feature is fixed Position effect significantly improves, and algorithm also reduces the requirement to image acquisition, improves the comfortableness of user, to gathered person simultaneously Finger stretching degree does not has rigors, and the user of defective to finger (bending, excalation) also is adapted for this algorithm.

The above, only the invention preferably detailed description of the invention, but the protection domain of the invention is not Being confined to this, any those familiar with the art is in the technical scope that the invention discloses, according to the present invention The technical scheme created and inventive concept thereof in addition equivalent or change, all should contain the invention protection domain it In.

Claims (4)

1. the palmmprint ROI dividing method in a personal recognition, it is characterised in that: comprise the steps:

S1. fitting a straight line is chosen；

S2. the ROI segmentation of image flame detection and palmmprint.

2. the palmmprint ROI dividing method in personal recognition as claimed in claim 1, the steps characteristic of step S1 is as follows: at figure Choosing the reference direction that a stable straight line is split as ROI in Xiang, the marginal point for profile diagram first quartile uses Little square law simulates straight line.

3. the palmmprint ROI dividing method in personal recognition as claimed in claim 1, the steps characteristic of step S2 is as follows: determine The central point of ROI, with valley point M1 as fixing point, finds on a M2 is expert at and makes straight line M1M2 become fixed angles with fitting a straight line L Point M2 ', the midpoint O of line taking section M1M2 ' of degree, does the perpendicular bisector of straight line M1M2 ', and in the right side area of perpendicular bisector Find the some O1 of regular length R, then some O1 is just in the central area of palm, and the finally central point with an O1 as ROI intercepts The square area of 128 × 128 is as the ROI of image, describedDescribed foursquare two of which limit is put down Row is in fitting a straight line L.

4. the palmmprint ROI dividing method in personal recognition as claimed in claim 2, it is characterised in that the step of fitting a straight line It is:

If the equation expression formula of straight line is:

Y=kx+b (1)

Measured value according to volar edge profile obtains straight line intercept b on the y axis and straight slope k, (x_i,y_i) it is palm limit The coordinate of the measured value of edge profile, b₀、k₀For the approximation of b, k, make:

B=b₀+δb

K=k₀+δk

Wherein, δ b and δ k is the deviation of slope and intercept；

Using y as dependent variable, with x as independent variable, error equation is:

$v_{yi}=\left[\begin{array}{cc}{x}_i& 1\end{array}\right]\left[\begin{array}{c}\mathrm{\δ}b\\ \mathrm{\δ}k\end{array}\right]+(b_0{x}_i+k_0-y_i)$

The matrix expression of error equation is:

A δ X=L+V

Wherein

$A=\left[\begin{array}{cc}{x}_1& 1\\ x_2& 1\\ \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}.\\ .\\ .\end{array}\\ x_n& 1\end{array}\right],L=\left[\begin{array}{c}{b}_0{x}_1+k_0-y_1\\ b_0{x}_2+k_0-y_2\\ \begin{array}{c}.\\ .\\ .\end{array}\\ b_0{x}_n+k_0-y_n\end{array}\right]$

$V=\left[\begin{array}{c}{v}_{y1}\\ v_{y2}\\ \begin{array}{c}.\\ .\\ .\end{array}\\ v_{yn}\end{array}\right],\mathrm{\δ}X=\left[\begin{array}{c}\mathrm{\δ}b\\ \mathrm{\δ}k\end{array}\right]$

By method of least square criterion

V^TV=min

I.e.

$\underset{i=1}{\overset{n}{\Σ}}\left|\right|{\mathrm{bx}}_i-k-y_i|{|}^2=min$

Its least square solution is:

$\mathrm{\δ}\hat{X}={\left(A^{T}A\right)}^{-1}{A}^{T}L---\left(2\right)$

The value drawing k, b with this, brings formula 1 into and i.e. obtains fit equation and fitting a straight line.