CN104112125A - Method for identity recognition based on palm print and finger crease feature fusion - Google Patents

Abstract

The invention relates to a method for identity recognition based on palm print and finger crease feature fusion. The method is achieved based on fusion of palm print and finger crease wavelet transform feature matching on a score layer. The method comprises the steps of adopting wavelet transform to perform feature extraction on palm print and finger crease images, and constructing palm print and finger crease wavelet energy characteristics; implementing corresponding feature matching operation on two wavelet energy features, obtaining respective initial matching scores and performing normalization processing; finally using multiplication rules to perform fusion on palm print and finger crease normalization matching scores, and obtaining a final fusion score. Recognition precision of the invention can be worked out based on statistical results of the fusion score, and accordingly effectiveness of the method is verified.

Description

Personal identification method based on palmmprint and finger band Fusion Features

Technical field

The present invention relates to a kind of new method of human body biological characteristics recognition technology, particularly a kind of fusion recognition new method based on palmmprint and finger band wavelet energy feature, belongs to computer application field.

Background technology

Human body biological characteristics recognition technology is a kind of personal identification technology that computing machine combines with high-tech means such as optics, acoustics and biostatistics principles, and it utilizes the intrinsic biological characteristic of human body to carry out personal identification evaluation.At present, human body identity identifying technology mainly adopts people's face, fingerprint, iris, palmmprint and refers to the features such as band.This category feature has plurality of advantages: (1) ubiquity, and anyone has this category feature; (2) uniqueness, has at everyone under the prerequisite of same feature, and the feature that any two people have is different; (3) permanent, this category feature has permanent unchangeability, can be not in time or the variation of environmental factor and large variation occurs; (4) security, this category feature is difficult for being forged or imitating; (5) collection property, feature can be collected easily; (6) acceptability, the feature of using should be accepted by user than being easier to; (7) performance requirement, is used the human body identity authorization system of this type of characteristic Design can obtain higher accuracy of identification.

Palmmprint identification is emerging in recent years a kind of personal identification method, is the important supplement to existing biometrics identification technology.Compare with other biological feature, palmmprint has following characteristics: (1), with people's appearance ratio, palmmprint gatherer process is easy to control, and has avoided human face expression to change the precision causing and has lost problem; (2) with fingerprint, refer to that band compares, palmmprint contains the more property the distinguished characteristic information of horn of plenty, and more holds susceptible to user acceptance; (3) with iris, compare, palmmprint collecting device is cheap, easy to use.Palmmprint has main line, fold and mastoid process line etc. and enriches streakline feature, mastoid process line more carefully, a little less than, can from high resolving power, high-quality image, extract, and main line, fold more slightly, stronger, can from low resolution, noisy image, extract.In addition, the main line in palmmprint and THE FOLD FEATURES also have different directional informations.

Equally, refer to that band recognition methods is as the important supplement of another kind of biometrics identification technology, it has certain intrinsic advantage equally.Refer to that band compares with people's face, iris, have that gatherer process is simple, collecting device is cheap, consumers' acceptable degree high; Finger band is compared with fingerprint, palmmprint, refers to that band, in finger bending place, is not generally prone to the loss of significance causing because of callus, dirt, and because long-term finger bending motion makes its feature more obvious, for identification provides larger convenience; Finally, refer to that band is distributed in different fingers upper, be more difficult for the imitation that is stolen.

Wavelet transformation, can be from different scale, diverse location and different directions analyzing and processing image texture information as a kind of multiresolution analysis instrument.For more slightly or stronger feature, can adopt the wavelet transformation under large scale, low resolution; Feature for more carefully or, can adopt small scale, high-resolution wavelet transformation.With wavelet transformation, carry out respectively palmmprint and refer to that band identification is a kind of known method.But utilize single palmmprint and finger band recognition technology to carry out living things feature recognition, its accuracy of identification still has the space of lifting.In addition, palmmprint and refer to that band feature is in together among palm utilizes its same equipment to gather when collection apparatus simultaneously, it is carried out to fusion recognition very convenient.

Summary of the invention

The object of the invention is to provide a kind of based on palmmprint and the method that refers to that band wavelet energy Fusion Features carries out authentication, and it can improve, and the single palmmprint based on wavelet transformation is identified or the accuracy of identification of the identification of the finger band based on wavelet transformation.

To achieve these goals, the technical solution adopted in the present invention is a kind of method of palmmprint and finger band feature that merges in fractional layer based on wavelet transformation; It is divided into palmmprint, refer to band feature extraction and fractional layer Fusion Features and mate two stages; Obtaining the palmmprint that pre-service is good and referring under the prerequisite of band image, from palmmprint and finger band image, extract respectively palmmprint wavelet energy feature and refer to band wavelet energy feature, calculate respectively both in the score of fractional layer, again palmmprint and finger band score are formed to new score according to fractional multiplication fusion method, use new score to identify individual; Detailed process is as follows:

The first step: palmmprint and the feature extraction of finger band wavelet energy;

(1) first palmmprint and finger band image are carried out to wavelet decomposition;

Decompose low frequency subgraph and the details subgraph obtain palmmprint, to refer to band, remove palmmprint, refer to the low frequency subgraph of band;

(2) each details subgraph is divided to mutually disjoint sub-block;

(3) calculate respectively palmmprint and refer to the corresponding energy of sub-block in each details subgraph of band, and constructing respectively palmmprint and refer to band proper vector;

Second step: characteristic matching and fractional layer merge;

(1) calculate respectively palmmprint and refer to band matching score, being divided into true matching score and false matching score, and carrying out method for normalizing processing;

(2), according to multiplication fusion method, in fractional layer, merge palmmprint and refer to band score.

Described wavelet decomposition is 3 grades, and in palmprint feature extraction, wavelet decomposition is used ' Haar ' small echo, refers to that in band feature extraction, wavelet decomposition is used ' Db1 ' small echo.

It is 12 that described palmmprint details subgraph divides line direction, and column direction piece is several to be calculated according to details subgraph size; Refer to that band details subgraph line direction is 3, column direction piece is several to be calculated according to details subgraph size.

Described details subgraph is divided into water horizontal direction details subgraph, vertical direction details subgraph and diagonal details subgraph.

Described line and finger band matching score adopt absolute distance computing method.

Described method for normalizing is z-score method for normalizing, and its formula is

Score＝(x-mean(x))/std(x)；

Wherein, x is the coupling score value vector between a class biological characteristic, and mean (x) and std (x) are palmmprint or mean value and the standard deviation that refers to the true and false coupling of band.

The formula of described multiplication fusion method is

score＝f _i*p _i(i＝1,2,...,M)；

Wherein, f _ifor referring to i score of band, p _ifor i score of palmmprint; M is total points order.

The present invention compared with prior art, has the following advantages:

(1) general palm grain identification method and the recognition methods of finger band based on small echo all only utilizes single palm print characteristics or refers to band feature, be subject to palmmprint and refer to band separately intrinsic characteristic information enrich the restriction of degree, the accuracy of identification that single method is obtained still has greater room for improvement; The present invention gathers palmmprint simultaneously and refers to band feature, takes full advantage of characteristic information separately, merges after utilizing small wave converting method to extract feature in fractional layer, and accuracy of identification improves a lot.

(2) method that biological characteristic merges has many kinds, such as the fusion of people's face and palm print characteristics, and fusion of people's face and fingerprint characteristic etc.But this type of all needs many complete equipments to gather respectively different human body feature.The feature that the present invention uses is in together among palm, only needs a set of equipment can complete the collection of feature during collection, has certain convenience, has reduced financial cost.

(3) the present invention, further by wavelet details subgraph piecemeal, calculates respectively the energy of palmmprint piece and finger band piece, takes full advantage of the local message of palmmprint and finger band image.

The new method that the palmmprint based on wavelet transformation that the present invention proposes and finger band merge, has certain validity, and its ratio of precision single method all increases.Palmmprint based on wavelet transformation identify its etc. error rate be 2.46%, and finger band based on wavelet transformation identify its etc. error rate be 5.62%, utilize that new method of the present invention obtains etc. error rate be 1.75%.

The present invention is elaborated with reference to accompanying drawing in connection with embodiment, to object of the present invention, and feature and carry out a little deep understanding.

Accompanying drawing explanation

Fig. 1 new method FB(flow block) of the present invention;

The result comparison of Fig. 2 the inventive method and single method.

Embodiment

As shown in Figure 1, the present invention is palmmprint based on wavelet transformation and refers to the method that band fractional layer merges, and it is divided into two stages: palmmprint, refer to band wavelet energy feature extraction phases and fractional layer Fusion Features.In order to verify validity of the present invention, we use the palmmprint of The Hong Kong Polytechnic University and refer to band database.

Specific implementation process is as follows:

The first step: the palmmprint based on wavelet transformation, refer to the feature extraction of band wavelet energy;

(1) the palmmprint energy feature based on wavelet transformation calculates

With different scale respectively in the horizontal direction, decompose palmprint image in vertical direction and diagonal, the wavelet coefficient obtaining on not at the same level and different directions has formed palm print characteristics to wavelet transformation.If H _i, V _iand D _ibe respectively i level small echo in the horizontal direction, the detail pictures obtaining of decomposing palmprint image in vertical direction and diagonal.In order to describe the basic building block of palmmprint texture, as palmmprint line and mastoid process line etc., we calculate respectively H _i, V _iand D _ithe wavelet energy of direction, represents each basic building block strength information in different directions with this.Palmprint image is at H _i, V _iand D _ii level wavelet energy computing formula in direction is as follows:

$E_i^h=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}{\left[H_i(x,y)\right]}^2---\left(1\right)$

$E_i^v=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}{\left[V_i(x,y)\right]}^2---\left(2\right)$

$E_i^d=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}{\left[D_i(x,y)\right]}^2---\left(3\right)$

Small echo is when decomposing non-oscillatory signal, and wavelet coefficient can form increase trend with the continuous increase of the progression decomposing; But when decomposing oscillator signal, oscillator signal is being far smaller than the wavelet coefficient of the wavelet decomposition level corresponding with its oscillation frequency on the contrary compared with the wavelet coefficient of high de-agglomeration level.Because palmmprint line does not have concussion, so the wavelet energy of palmmprint line mainly concentrates in the wavelet decomposition detail image of large scale.And mastoid process line has certain concussion, therefore the wavelet energy of mastoid process line mainly concentrates on compared with in the wavelet details image of small scale.Based on this, the proper vector consisting of with this wavelet energy in different wavelet decomposition level can be expressed as:

${(E_i^h,E_i^v,E_i^d)}_{i,2,...,M}---\left(4\right)$

Wherein, M is total progression of wavelet decomposition.

(2) the finger band energy feature based on wavelet transformation calculates

Refer to band wavelet energy feature calculation process and palmmprint wavelet energy similar computation process, repeat no more herein.

(3) structure of palmmprint and finger band feature

From formula (4), can find out, be the global information of having described respectively palmmprint and having referred to band image by this proper vector advantage of wavelet energy structure, and shortcoming is the local message not having in Description Image.Therefore, this feature is as the good textural characteristics of picture engraving of a kind of global characteristics.For local feature that can better Description Image texture, each width detail pictures that we obtain wavelet decomposition is divided into respectively S ₁* S ₂individual mutually disjoint, refer to that its method of partition of band image and palmprint image are similar.Then, calculate respectively the wavelet energy of every.Finally, with calculating these non-intersect energy that obtain, re-construct following palmmprint textural characteristics vector :

$\tilde{V}=({\tilde{V}}_{\left(1\right)}^1,{\tilde{V}}_{\left(2\right)}^1,...,{\tilde{V}}_{(3\×S_1\×S_2)}^1,...,{\tilde{V}}_{\left(1\right)}^M,{\tilde{V}}_{\left(2\right)}^M,...,{\tilde{V}}_{(3\×S_1\×S_2)}^M)---\left(5\right)$

Wherein, M is the small echo maximum decomposition level number of setting, by i level wavelet decomposition H _i, V _iand D _ithe detail view that 3 directions obtain is divided into S ₁* S ₂the energy of each piece calculating after individual mutually disjoint.

Finally, right be normalized according to the following formula:

$V_{\left(j\right)}^i=\frac{{\tilde{V}}_{\left(j\right)}^i}{\underset{k=1}{\overset{M}{\mathrm{\Σ}}}\underset{l=1}{\overset{3\×S_1\×S_2}{\mathrm{\Σ}}}{\tilde{V}}_{\left(1\right)}^K}(i=1,...,M;j=1,...,3\×S_1\×S_2)---\left(6\right)$

$V=(V_{\left(1\right)}^1,V_{\left(2\right)}^1,...,V_{(3\×S_1\×S_2)}^1,...,V_{\left(1\right)}^M,V_{\left(2\right)}^M,...,V_{(3\×S_1\×S_2)}^M)---\left(7\right)$

The V obtaining after normalization is corresponding to palmmprint and refer to that band is called as wavelet energy feature.I level wavelet energy feature can according to as give a definition:

$V^i=(V_{\left(1\right)}^i,V_{\left(2\right)}^i,...,V_{(3\×S_1\×S_2)}^i)---\left(8\right)$

Bringing above formula into formula (7) can obtain:

V＝(V ¹,V ²,...,V ^M) (9)

I level wavelet energy feature is calculated by i level coefficient of wavelet decomposition, and it has reflected that image is at yardstick 2 ^-iupper, the texture features on diverse location and different directions.And above formula is to be combined successively by wavelet energy features at different levels, reflected the texture features of image on different scale, diverse location and different directions.To sum up, palmmprint wavelet energy characteristic extraction procedure of the present invention can be summed up as following 4 steps:

A) carrying out 3 grades of decomposition through pretreated palmprint image subgraph;

B) each palmmprint details subgraph is divided into S ₁* S ₂individual mutually disjoint sub-block, each details subgraph line direction piece number of palmmprint is 12, the piece of column direction is several according to the calculating of palmmprint details subgraph size;

C) each palmmprint details subgraph is calculated to the wherein energy of each sub-block, and be configured to according to this vector;

D) normalized vector, obtains the palmmprint wavelet energy feature for mating.

Middle finger band wavelet energy characteristic extraction procedure of the present invention and palmmprint wavelet energy characteristic extraction procedure are similar, and difference is b) the piece number of step middle finger band image row direction is 3, the piece of column direction is several according to referring to that band details subgraph size calculates.

Second step: characteristic matching and fractional layer merge

Through palmmprint with refer to after band feature extraction, use following absolute value distance formula to calculate two palmmprints or refer to the similarity of band:

$D_{\mathrm{ij}}=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}|V_i-V_j|---\left(10\right)$

Wherein, D _ijthe similarity that represents an i and j proper vector.Then carry out true coupling and false coupling counts the score, the score that comes from the homology palmmprint of same person or refer to band is less than allos palmmprint or refers to the score of band.Finally, palmmprint and finger band score are normalized, normalization formula is as follows:

$s=\frac{x-\mathrm{mean}\left(x\right)}{\mathrm{std}\left(x\right)}---\left(11\right)$

Wherein, mean (x) and std (x) are palmmprint or mean value and the standard deviation that refers to the true and false coupling of band.Use new score merge palmmprint and refer to band score according to multiplication fusion formula, its formula is as follows:

score＝f _i*p _i(i＝1,2,...,M) (12)

Wherein, f _ifor referring to i score of band, p _ifor i score of palmmprint; M is total points order.Use new score score to calculate accuracy of identification, the error rate such as use to weigh accuracy of identification herein, new method of the present invention and single palmmprint or refer to band accuracy of identification more as shown in Figure 2.

In a word, being a kind of palmmprint based on wavelet transformation and referring to that band fractional layer merges new method of above-mentioned explanation and elaboration, it is applicable to palmmprint and refers to that band feature is in the fusion of fractional layer, to improve the single palm print characteristics based on wavelet transformation or to refer to the accuracy of identification of band feature extraction.Preferred example of the present invention is set forth, in this example, the present invention obtain etc. error rate be 1.75%, the error rates 5.62% such as the error rates such as more single palm print characteristics 2.46% and single finger band feature have a distinct increment.

Claims (6)

1. based on palmmprint with refer to the personal identification method of band Fusion Features, it is characterized in that: it is divided into palmmprint, refer to band feature extraction and fractional layer Fusion Features and mate two stages; Obtaining the palmmprint that pre-service is good and referring under the prerequisite of band image, from palmmprint and finger band image, extract respectively palmmprint wavelet energy feature and refer to band wavelet energy feature, calculate respectively both in the score of fractional layer, again palmmprint and finger band score are formed to new score according to fractional multiplication fusion method, use new score to identify individual; Detailed process is as follows:

(1) first palmmprint and finger band image are carried out to wavelet decomposition;

Decompose low frequency subgraph and the details subgraph obtain palmmprint, to refer to band, remove palmmprint, refer to the low frequency subgraph of band;

(2) by palmmprint, refer to that each details subgraph of band divides mutually disjoint sub-block;

(3) calculate respectively palmmprint and refer to the corresponding energy of sub-block in each details subgraph of band, and constructing respectively palmmprint and refer to band proper vector;

(4) calculate respectively palmmprint and refer to band matching score, being divided into true matching score and false matching score, and carrying out method for normalizing processing;

(5), according to multiplication fusion method, in fractional layer, merge palmmprint and refer to band score.

2. according to right, want the personal identification method based on palmmprint and finger band Fusion Features described in 1, its feature is being: the wavelet decomposition in described step (1) is 3 grades, in palmprint feature extraction, wavelet decomposition is used ' Haar ' small echo, refers to that in band feature extraction, wavelet decomposition is used ' Db1 ' small echo.

According to right, want the personal identification method based on palmmprint and finger band Fusion Features described in 1, its feature is being: it is 12 that the palmmprint details subgraph in described step (2) divides line direction, and column direction piece is several to be calculated according to details subgraph size; Refer to that band details subgraph line direction is 3, column direction piece is several to be calculated according to details subgraph size.

3. according to right, want the personal identification method based on palmmprint and finger band Fusion Features described in 1, its feature is being: in described step (2), details subgraph is divided into water horizontal direction details subgraph, vertical direction details subgraph and diagonal details subgraph.

4. according to right, want the personal identification method based on palmmprint and finger band Fusion Features described in 1, its feature is being: in described step (2), line and finger band matching score adopt absolute distance computing method.

5. according to right, want the personal identification method based on palmmprint and finger band Fusion Features described in 1, its feature is being: in described step (2), method for normalizing is z-score method for normalizing, and its formula is

Score＝(x-mean(x))/std(x)；

Wherein, x is the coupling score value vector between a class biological characteristic, and mean (x) and std (x) are palmmprint or mean value and the standard deviation that refers to the true and false coupling of band.

6. according to right, want the personal identification method based on palmmprint and finger band Fusion Features described in 1, its feature is being: in described step (5), the formula of multiplication fusion method is

score＝f _i*p _i(i＝1,2,...,M)；

Wherein, f _ifor referring to i score of band, p _ifor i score of palmmprint; M is total points order.