WO2016011640A1

WO2016011640A1 - Identification method based on handprint imaging

Info

Publication number: WO2016011640A1
Application number: PCT/CN2014/082924
Authority: WO
Inventors: 徐勇; 颜珂; 汪晓丹
Original assignee: 哈尔滨工业大学深圳研究生院
Priority date: 2014-07-24
Filing date: 2014-07-24
Publication date: 2016-01-28

Abstract

An identification method based on handprint imaging, comprising the following steps: step I: registration, respectively acquiring a palm print and fingerprint; step II: image pre-processing; step III: determining whether an input image is a palm print or a fingerprint; step IV: palm print identification. The identification method abandons a traditional single identification mode, and comprises both palm print identification and fingerprint identification systems. A to-be-tested person inputs a palm print or fingerprint, and the system automatically identifies that the input characteristic is the palm print or the fingerprint, and finalizes matching. The present invention creatively introduces a double-fingerprint identification, thus further improving the precision of fingerprint identification.

Description

Identity recognition method based on hand image texture

TECHNICAL FIELD The present invention relates to an identification method, and more particularly to an identification method based on a hand image texture. BACKGROUND OF THE INVENTION Fingerprint is one of the earliest biological features used to identify identities. It consists of ridge lines and valley lines. It is unique and life-long for everyone, and thus becomes one of the most commonly used and most reliable means of identification. Pick your thumb or index finger for identification. Fingerprint recognition is widely used, and it is mainly used in the fields of attendance, customs, public security and security. Palmprint is a rising star in biometrics relative to fingerprints. It has similar characteristics to fingerprints. It consists of various kinds of lines, and it is also unique and lifelong, but it also has its distinctive features:

1. The palm print area is large, so more information can be provided; 2. In addition to the ridge line and the valley line, the palm print has other characteristics (lifeline, career line, wisdom line, etc.), and also has uniqueness and lifetime. Sexuality; 3, it is usually possible to obtain the contour information of the palm when acquiring the palm print; and so on. Based on the above characteristics of the fingerprint and the palm print, both of these features have the advantages of convenient collection and easy promotion. However, most of the market-based identification devices based on palm prints and fingerprints have a single function and can only recognize a single biological feature. If the subject's fingerprint or palm print is injured or inconvenient to identify, the identification cannot be performed. In view of the market gap and demand, this paper proposes a combination of palmprint and fingerprint reader. If the fingerprint is not convenient to identify, it can be identified by palmprint, and vice versa, which greatly increases the flexibility of recognition and introduces double fingerprints in a pioneering manner. The index finger and the middle finger are recognized, and the fingerprint recognition accuracy is improved, thereby improving the overall system recognition effect.

SUMMARY OF THE INVENTION In order to solve the problems in the prior art, the present invention provides an identification method based on hand image texture, which includes the following steps:

Step 1: Register, collect palm prints and fingerprints separately; Step 2: Image preprocessing;

Step 3: Determine whether the input image is a palm print or a fingerprint:

Detecting the ratio of effective pixels in the image to the entire image ^^, wherein the effective pixel refers to the pixel corresponding to the texture, and setting a threshold value of 73⁄4 ^^, if the effective pixel ratio is greater than a given threshold, determining that the image is a palm print image, According to the palmprint recognition method for identification, and vice versa, the fingerprint is identified according to the fingerprint identification method;

Step 4: The palmprint recognition steps are as follows:

Step 4. 1 : Extracting the palm contour

Edge detection is performed on the palmprint image, and the edge of the image is formed into a unimodal function by a differential operator, and the peak position corresponds to the edge point, and is extracted to the palmprint contour; Step 4. 2: Calculating the shape context uniformly selects M samples on the palmprint contour Point, given a point, and use it as the origin of the log polar coordinate system, the coordinates of other points are obtained based on the given point, connecting the point with the remaining points will get - 1 vector, these vectors can describe the palm The contour of the pattern is based on the position distribution of a given point. The transformation formula from Cartesian coordinates to log polar coordinates is as follows:

p = \ g ^x ² + y ² .

θ = arctan y I x (if x > 0) Step 4. 3: Palmprint contour matching transforms the matching problem of the test sample and the training sample based on the palmprint contour into a bipartite graph matching problem with weights, where the weight It is defined as the matching distance, that is, the matching distance between all matching pairs. It is assumed that there is a class C palm print, and each type of palm print has m training samples, X, = [ .., ], = l, 2, ... C For the first type of palmprint sample, y is the test sample, then the following method is used for palmprint recognition: For each training sample in each class, the perfect match between the test sample y and the palmprint contour matching weight is obtained, and the minimum is recorded. Matching weights: - arg m{w eighty {π _} ) } where ^. indicates the corresponding first perfect match, indicating the matching match corresponding to the contour Value, indicating the selected match,

According to the smallest matching weights from small to large, the training samples corresponding to the first N minimum matching weights are selected to construct a new training sample space Χ'={Α, ' ₂ ,..., ^} _;

Step 4.4: Feature Point Matching

Extract feature points from the palmprint image in the new sample space, and select the feature points closest to the palm print center to identify the match.

For the samples in the new sample space, find the perfect match between the test sample and its feature point matching weight, and record the minimum matching weight.

Φ _Ν = wgm {weight' _Ν (^)}

Where Α denotes the first perfect match, v^g ) denotes the matching weight corresponding to the feature point matching, and φ _Ν denotes the selected match,

When the sum of the minimum feature point matching weights of all the training samples of a certain class in the new sample space is the smallest, then the test sample is determined to belong to the class.

Class = gm {T weight^ {φ)}^ If it is a fingerprint, it can be identified according to the conventional method of the prior art or the following method.

As a further improvement of the present invention, the image preprocessing process of the second step includes: filtering, binarizing, refining, extracting feature points, and removing pseudo feature points.

As a further improvement of the present invention, refinement: extracting the centerline of the ridge to a pixel width instead of the ridge line without changing the fingerprint and the palm line topology; extracting the feature points: extracting the feature points, including : starting point, bifurcation point, intersection point; removing pseudo feature points: In the feature point extraction process, it is easy to misjudge the fingerprint and palm line at the edge of the image as the starting point. These false positive points are called pseudo Feature points, need to be removed.

As a further improvement of the present invention, in step 4.1, the palm contour is extracted.

The Laplace-Gaussian operator performs edge detection on the palmprint image. As a further improvement of the present invention, the matching algorithm in step 4.3 adopts a Hungarian match ₍ as a further improvement of the present invention, fingerprint recognition includes the following steps:

1) Fingerprint segmentation

Find the connection between the last two points of the two fingerprints and the midpoint of the connection. The midpoint is used as the normal of the connection. The normal fingerprint is used as the dividing line to divide the fingerprint image into the fingerprint image of the index finger and the fingerprint image of the middle finger. The following operations are performed on the two fingerprints. Fingerprint images are separately performed;

2) Gradient method for direction field estimation

Dividing the fingerprint image into small pieces of ννχνν;

Calculate the gradient 0', and the gradient in the y direction of each pixel ( ) in each small block ( );

Estimate the direction of each point

^2d _x {u,v)d _y {u,v)

1

∑ ∑3⁄4(α,ν)^(α,ν)

3) Detection core point

The core points are detected by curvature, and the X-direction component and the y-direction component difference are calculated for each small block:

Diff Y = ^ sin 20(k, w)- sin 20(k,l)

Diff X = ^cos2^(u-J) -^cos26»(l,/)

1=1 l=\ If the 0 Y and X on the point (^) are both negative, the point is the curvature point C,

Bit to the core point;

4) Registration

Given a fingerprint template, register all fingerprints based on the template fingerprint:

Where is the point on the original fingerprint image, Ατ, Δ^, Δ^ respectively represent the displacement difference and the angle difference between the original fingerprint image and the template image in the X direction, the y direction, and can be based on the original fingerprint image and the center point on the template image These three parameters are calculated;

5) Identification

Suppose there are C individuals who have collected double fingerprints. Each person has collected m training samples. X, = [ .., ], = l, 2, ... C indicates the type of double fingerprint samples, and y is the test sample. Use the following methods for fingerprint recognition:

Select the feature points closest to the fingerprint core point to identify the match;

After the two fingerprint block matching respectively make the category attribution decision, calculate the average error between each fingerprint block and all training samples of the category: 丄∑|<L„_ y| where w = l, represents the index finger fingerprint block , when w = 2, represents the middle finger fingerprint block, indicating the category decision of the corresponding block;

Produce the final decision:

Desision - argimnii As an advancement of the present invention, the gradient calculation is calculated using a template. The beneficial effects of the present invention are: The recognition method discards the conventional single recognition mode, including the palmprint recognition and fingerprint recognition system. The test subject can input any one of palm print and fingerprint, and the system will automatically recognize the input feature as palm print or fingerprint, and finally complete the match. Pioneering the introduction of dual fingerprint recognition to further improve fingerprint recognition accuracy.

DRAWINGS 1 is a schematic structural view of a palmprint and fingerprint recognition system of the present invention;

2 is a flow chart of the identification process of the present invention.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, the present invention will be further described with reference to the accompanying drawings.

Palmprint and fingerprint identification steps:

Step 1: Registration stage, collect palm prints and fingerprints

Step 2: Image Preprocessing Stage

Due to the dryness of the skin of the hand during collection and the noise interference generated by the collection device itself, the effects of the collected palm prints and fingerprint images are often not ideal, which affects the effect of subsequent feature point extraction. In order to eliminate noise, we need to preprocess palmprint and fingerprint images. The general preprocessing process includes: filtering, binarization, refinement, extracting feature points, and removing pseudo feature points.

Filtering: Filter out unwanted information (noise) and enhance useful information;

Binarization: Since palm prints and fingerprints only include ridges and valleys, they can be completely described by binarized images;

Refinement: Under the premise of not changing the fingerprint and palm line topology, the centerline of the line is extracted to a pixel width instead of the line, which is beneficial to the extraction of subsequent feature points.

Extract feature points: Extract detailed feature points, including: start and end points, bifurcation points, and intersection points.

Removal of pseudo-feature points: In the feature point extraction process, it is easy to misjudge the fingerprint and the palm line at the edge of the image as the end point. These misjudged points are called pseudo-feature points and need to be removed.

Step 3: Identification stage

Before we can identify, we need to judge whether the input image is palm print or fingerprint. The method used in this paper is to detect the ratio of effective pixels in the image to the whole image. , where the effective pixel refers to the pixel corresponding to the texture, and sets a threshold value of 73⁄4 ^^, if the effective pixel ratio is greater than a given threshold, then The image is broken as a palmprint image, and is identified according to the palmprint recognition method in the following text, and is determined as a fingerprint, and is identified according to the fingerprint recognition method. Input image = _{+ /} , ^..

L fingerprint image, RWz'i? < Threshold

1, palmprint recognition:

1) Extract the palm contour

We use the Laplace-Gaussian operator to perform edge detection on the palmprint image. The edge of the image is formed by a differential operator to form a unimodal function, and the peak position corresponds to the edge point. Based on this, we extracted the palm outline. 2) Calculate the shape context

Select M sample points evenly on the palmprint outline, give a point, and use it as the origin of the log polar coordinate system. The coordinates of other points are obtained based on the given point. Connecting the point with the remaining points will get - 1 vector, these vectors can describe the positional distribution of the palmprint outline based on a given point. From Cartesian coordinates

(Cartesian coordinates) The transformation formula to the log polar coordinates is as follows:

p = \ g^x ² + y ²

θ = arctan y I x (if x>0)

3) Palmprint contour matching

For the test sample and the training sample based on the palmprint contour matching problem can be transformed into a weighted bipartite graph matching problem, where the weight can be determined as the matching distance, that is, the matching distance between all matching pairs. The matching algorithm in this paper takes a Hungarian match.

Suppose there is a class C palm print, each type of palm print has m training samples, X, =[ .., ], = l,2,...C represents the first type of palmprint sample, and y is the test sample, then the following The method performs palmprint recognition.

For each training sample in each class, the perfect match between the test sample y and the palmprint contour matching weight is obtained, and the minimum matching weight is recorded. - arg mm{w eighty {π . ) } Where ^. indicates the corresponding first perfect match, indicating that the contour matches the corresponding matching weight, indicating the selected match.

According to the smallest matching weights from small to large, the training samples corresponding to the first N minimum matching weights are selected to construct a new training sample space Χ'={Α, ' ₂ , . . . , ^}.

4) Feature point matching

The feature points are extracted from the palmprint image in the new sample space, and the feature points closest to the palmprint center are selected for identification matching. The matching algorithm still selects the Hungarian algorithm.

For the samples in the new sample space, the perfect match between the test sample and its feature point matching weight is obtained, and the minimum matching weight is recorded.

Φ _Ν = wgm {weight' _Ν (^)} where Α represents the first perfect match,

) indicates that the feature points match the corresponding matching weights, and φ _Ν indicates the selected matches.

Class - arg min{ V weighty ( )}

2, fingerprint identification:

1) Fingerprint segmentation

Find the line connecting the last two points between the two fingerprints and the midpoint of the connection. The midpoint is used as the normal of the connection. The normal fingerprint is used as the dividing line to divide the fingerprint image into the index finger image and the middle finger fingerprint image. The following operations are performed separately for the two fingerprint images.

2) Direction field estimation

There are many ways to estimate the directional field. This paper uses the most widely used gradient method.

1 Split the fingerprint image into small pieces of W 2 Calculate the gradient 0·, j') in the χ direction of each pixel in each small block, and the gradient 0·, ·) in the y direction. (Gradient calculation can be calculated using a template)

3 Estimate the direction of each point

∑ ∑2d _x (u,v)d _y (u,v)

, ∑ ∑3⁄4(a,v)aj(a,v)

3) Detection core point

The core points are detected by curvature, and the X-direction component and the y-direction component difference are calculated for each small block separately.

Diff Y = Jsin 20(k, w)-^ sin 20(k,l)

Diff X = ^ cos 20(w ) - ^ cos 26»(1, /) If Diff Y and Diff X on the point (ί, j') are both negative, then the point is the curvature point C, generally w = 3 You can locate the core point.

4) Registration

Where is the point on the original fingerprint image, Ατ, Δ^, Δ^ respectively represent the displacement difference and the angle difference between the original fingerprint image and the template image in the X direction, the y direction, and can be based on the original fingerprint image and the center point on the template image These three parameters are calculated.

6) Identification

Suppose there are C individuals who have collected double fingerprints. Each person has collected m training samples. X, =[ .., ], = l,2,...C indicates the type of double fingerprint samples, and y is the test sample. Use the following methods for fingerprint recognition

Select the feature points closest to the core point of the fingerprint to identify the match, and the matching algorithm selects the Hungarian tooth; The algorithm is similar to the palmprint feature point matching, and will not be described here.

After the two fingerprint block matches have made the category attribution decision, we calculate the average error between each fingerprint block and all training samples of the category: 丄∑|<L„ _ y| where w = l, representing the index finger fingerprint Block, w = 2, represents the middle finger fingerprint block, indicating the category decision of the corresponding block.

Produce the final decision:

Desision - argimnii ^(lv) . The above is a detailed description of the present invention in conjunction with the specific preferred embodiments, and the specific embodiments of the present invention are not limited to the description. It will be apparent to those skilled in the art that the present invention may be made without departing from the spirit and scope of the invention.

Claims

Claim

A method for identifying an image based on a hand image texture, comprising the following steps: Step 1: registering, respectively collecting palm prints and fingerprints;

Step 2: Image preprocessing;

Step 3: Determine whether the input image is a palm print or a fingerprint:

Detecting the ratio of effective pixels in the image to the entire image ^^, where the effective pixel refers to the pixel corresponding to the texture, and sets a threshold value of 73⁄4^^. w, if the effective pixel ratio is greater than a given threshold, the image is judged to be a palmprint image, and is identified according to the palmprint recognition method, and the fingerprint is determined as a fingerprint, and the fingerprint recognition method is used for recognition;

Step 4: The palmprint recognition steps are as follows:

Step 4. 1 : Extracting the palm contour

Edge detection is performed on the palmprint image, and the edge of the image is formed into a unimodal function by a differential operator, and the peak position corresponds to the edge point, and is extracted to the palmprint contour; Step 4. 2: Calculating the shape context uniformly selects w samples on the palmprint contour Point, given a point, and use it as the origin of the log polar coordinate system, the coordinates of other points are obtained based on the given point, connecting the point with the remaining points will get -1 vector, these vectors can describe the palm The contour of the pattern is based on the position distribution of a given point. The transformation formula from Cartesian coordinates to log polar coordinates is as follows:

p = log^x ² + y ² .

θ = arctan ylx (if x > 0) Step 4. 3: Palmprint contour matching transforms the matching problem between the test sample and the training sample based on the palmprint contour into a bipartite graph matching problem with weights, where the weight is determined as Matching distance, that is, all matching pair distances after matching, assuming a class C palm print, each type of palm print has m training samples, = [«,..., ], i = l, 2,...C For the first type of palmprint sample, y is the test sample, then the following method is used for palmprint recognition: For each training sample in each class, the test sample y is matched with the palmprint contour matching weight. The smallest perfect match, record its minimum matching weight: π" = arg min { weight^ π _} ) } where ^ denotes the corresponding first perfect match, w gM^. ) denotes the matching weight corresponding to the contour match, r denotes The selected matching is sorted according to the minimum matching weight from small to large, and the training samples corresponding to the first N minimum matching weights are selected to construct a new training sample space Χ '= {« ,..., ^ } _; 4: Feature point matching extracts feature points from the palmprint image in the new sample space, and selects the feature points closest to the palm print center to identify the match. For the samples in the new sample space, the test sample is matched with its feature points. The perfect match with the smallest weight, record its minimum matching weight,

Φ _Ν = arg min{ wez'g zi ( ι ) } where A denotes the first perfect match, vm'gfe' ) denotes the matching weight corresponding to the feature point match, φ _Ν denotes the selected match, when the test sample and the new sample When the sum of the minimum feature point matching weights of all training samples of a certain type in space is the smallest, it is determined that the test sample belongs to the class.

Class - arg min{ ^ weight^ (φ)}.

2. The hand image texture based identification method according to claim 1, wherein: the image preprocessing process of the second step comprises: filtering, binarizing, refining, extracting feature points, and removing pseudo feature points. .

3. The hand image texture based identification method according to claim 2, wherein: the thinning: extracting the center line of the line to a pixel width without changing the fingerprint and the palm line topology Instead of lines; extract feature points: extract detailed feature points, including: start point, bifurcation point, cross point; remove pseudo feature points: in the feature point extraction process, it is easy to put the fingerprint and palm line on the edge of the image Partial misjudgment is the starting point. These points of misjudgment are called pseudo feature points and need to be removed.

4. The hand image texture based identification method according to claim 1, wherein: in step 4.1, the palmprint contour is extracted by using a Laplace-Gaussian operator to perform edge detection on the palmprint image.

5. The hand image texture based identification method according to claim 1, wherein: the matching algorithm in step 4.3 adopts a Hungarian matching.

6. The hand image texture based identification method according to claim 1, wherein: the fingerprint identification comprises the following steps:

2) Fingerprint segmentation finds the connection between the last two points of the two fingerprints and the midpoint of the connection. The midpoint is used as the normal of the connection. The normal fingerprint is used as the dividing line to divide the fingerprint image into the index finger image and the fingerprint image of the middle finger. The following operations are performed separately on two fingerprint images;

2) Gradient method for direction field estimation to segment the fingerprint image into small blocks of ν χ ; Calculate the gradient d _x i, j) in the X direction and the gradient in the y direction for each pixel ( j) in each small block Estimate the direction of each point: , ) = itan- ¹ ;

3) Detecting the core points The core points are detected by the curvature, and the difference between the directional component and the y-direction component is calculated for each small block:

Diff Y = ^sin 20{k, w) - ^ sin 2 , 1)

k = l k=l

Diff X = ^ cos 2^( w, ) - ^ cos 2Θ(\, I) If the point on the £> ¥ and £> X is negative, then the point is the curvature point C, generally w = 3 can be positioned To the core point; Registration is given a fingerprint template, and all fingerprints are registered based on the template fingerprint:

Where is the point on the original fingerprint image, Δ^, Δ^ respectively represent the displacement difference and the angular difference between the original fingerprint image and the template image in the direction and direction, which can be based on the original fingerprint image and the center point on the template image. Parameter calculation; recognition assumes that C individuals have collected double fingerprints, each person has collected m training samples, = [«,..., ], i = l, 2,...C indicates the class of double fingerprint samples For the test sample, the following methods are used for fingerprint recognition: Select the feature points closest to the fingerprint core point to identify the match; when the two fingerprint block matches respectively make the category attribution decision, calculate each fingerprint block and belong to it. Average error between all training samples in the category: d \

m ' 〉 y where _{w =} l, represents the index finger fingerprint block, w = 2, represents the middle finger fingerprint block, indicating the category decision of the corresponding block; the final decision is made:

o)

Desision - argimnJ

7. The hand image texture based identification method according to claim 6, wherein: in step 2), the gradient calculation is performed using a So Z template.