CN100550037C - Utilize and improve Hausdorff apart from the method for extracting the identification human ear characteristic - Google Patents

Abstract

The present invention relates to a kind of the utilization and improve Hausdorff apart from the method for extracting the identification human ear characteristic.The present invention comprises denoising, bulk normalization and the illumination compensation etc. of the collection of ear image, non-colour of skin noise by the pre-service to ear image, obtains the standard ear image.Adopt the method for gray scale morphology gradient and local Threshold Segmentation to extract people's edge feature in one's ear again, obtain normal man's edge image in one's ear.Adopt the inventive method, by standard variance and the improved Hausdorff distance of the intersegmental length difference of edge line, reduced the influence of the non-contour edge line segment point of point set (outfield point), obtain noiseproof feature preferably, strengthened based on Hausdorff and be used for the accuracy of people's edge image recognition in one's ear, made that the discrimination to people's ear improves greatly apart from the eigenwert that obtains.

Description

Utilize and improve Hausdorff apart from the method for extracting the identification human ear characteristic

Technical field

The invention belongs to person identification technology, particularly utilize and improve Hausdorff apart from the method for extracting the identification human ear characteristic based on human body biological characteristics.

Technical background

The ear recognition technology is a kind of biometrics identification technology that late 1990s begins to rise.The unique physiological characteristic that people's ear has and the advantage of observation angle make the ear recognition technology have suitable theoretical research value and actual application prospect.People's external ear divides auricle and external auditory meatus, and the object of ear recognition is actually the exposed auricle of external ear, just people's said traditionally " ear ".The complete people's ear automatic recognition system of one cover generally comprises following process: the identification of the cutting apart of the pre-service of ear image collection, image, ear image, feature extraction and ear image.

Hausdorff distance is usually used in weighing two similarity degrees between the somes set as a kind of distance measure, than binaryzation correlation technique better robustness is arranged for the undulatory property of the position of pixel.Owing to need not the corresponding relation between points of considering that two points are concentrated when using the Hausdorff distance, therefore can solve the identification problem when having noise in the image effectively as distance measure.In addition, Hausdorff does not need to carry out complicated optical flow computation or image segmentation coupling apart from the method for template matches.Just because of above-mentioned good characteristic, it often is used to realize that identification and tracking to the general objectives object detect, as terrain match, infrared remote sensing images match, Recognition of License Plate Characters etc.

Same, Hausdorff is apart from the middle of the identification that also can be used for people's ear.Because the edge of people's ear is to illumination-insensitive, people's relative being not easy of edge in one's ear is subjected to the influence of extraneous illumination variation when gathering ear image with digital equipment.So, can be bianry image (0,1 expression) with the edge transition of the ear image that obtains, obtain " normal man is the edge image in one's ear " after treatment, and adopt the Hausdorff distance to carry out ear recognition.But because the Hausdorff distance is a minimax distance, so it still is very sensitive to the noise in the ear image.In ear recognition, although may be two closely similar even identical people edge images in one's ear, because the existence of some false contourings (inhuman ear main outline line segment) also can make the calculating of Hausdorff distance produce very big error.And rim detection itself is exactly a uncertain problem, and the edge definition is normally very fuzzy, and it is very difficult to distinguish marginal point and noise spot.Therefore this has also had influence on the calculating of Hausdorff distance, thus need improve traditional Hausdorff distance, with raising based on people's ear recognition rate of edge characteristic information in one's ear.

Summary of the invention

The purpose of this invention is to provide a kind of the utilization and improve Hausdorff apart from the method for extracting the identification human ear characteristic.Adopt the inventive method, by standard variance and the improved Hausdorff distance of the intersegmental length difference of edge line, reduced the influence of the non-contour edge line segment point of point set (outfield point), obtain noiseproof feature preferably, strengthened based on Hausdorff and be used for the accuracy of people's edge image recognition in one's ear, made that the discrimination to people's ear improves greatly apart from the eigenwert that obtains.

Technology contents of the present invention is as follows:

Ear image is carried out pre-service, comprise denoising, bulk normalization and the illumination compensation etc. of the collection of ear image, non-colour of skin noise, utilize method to obtain people's edge image in one's ear based on gray scale morphology gradient and local Threshold Segmentation.Improve original Hausdorff distance algorithm with standard variance and the intersegmental length difference of edge line, be used to describe people's characteristic information of edge image in one's ear, concrete step is as follows:

(1) ear image of gathering is carried out pre-service:

By denoising and reparation to the non-colour of skin noise of the ear image gathered, adopt bulk normalization and illumination compensation, obtain the standard ear image;

(2) ear image Edge Gradient Feature:

The method that all gray scale morphology gradients and local Threshold Segmentation combine is extracted people's feature of edge in one's ear, obtains normal man's edge image in one's ear;

(3) utilize the intersegmental length difference of standard variance and edge line to improve the Hausdorff distance;

(4) with the people that obtained in the step (2) in one's ear its main edge contour of edge image be divided into 4 feature line segments of ear recognition, determine its human ear characteristic vector with improved Hausdorff distance.

Based on standard deviation and the improved Hausdorff distance of the intersegmental length difference of edge line (Length﹠amp; StandardDeviation Modified Hausdorff Distance, LSDMHD) with existing Hausdorff distance (Hausdorff Distance, HD), average Hausdorff distance (Modified HausdorffDistance, MHD) compare, concerning the Hausdorff distance, given two finite point set A={a ₁, a ₂..., a _mAnd B={b ₁, b ₂..., b _n, the Hausdorff distance definition between them is as follows:

H(A，B)＝max[h(A，B)，h(B，A)]

Wherein:

$h(A,B)=\underset{a\∈A}{\max }\underset{b\∈B}{\min }\left|\right|a-b\left|\right|$

$h(B,A)=\underset{b\∈B}{\max }\underset{a\∈A}{\min }\left|\right|b-a\left|\right|$

But because the Hausdorff distance is a maximum or minor increment, so it is very sensitive to the noise in the image.When being used for ear recognition, although obtain two closely similar objects, because the existence of some false contourings may make the calculating of Hausdorff distance produce very big error.

Average Hausdorff has obtained noiseproof feature preferably apart from having adopted the method for calculating mean distance, has reduced the influence of point set outfield point.

$h_{\mathrm{MHD}}(A,B)=\frac{1}{N_A}\underset{a\∈A}{\mathrm{\Σ}}\underset{b\∈B}{\min }\left|\right|a-b\left|\right|$

$h_{\mathrm{MHD}}(B,A)=\frac{1}{N_B}\underset{b\∈B}{\mathrm{\Σ}}\underset{a\∈A}{\min }\left|\right|b-a\left|\right|$

Above, N _AThe number of expression point set A mid point; N _BThe number of expression point set B mid point.In their substitution Hausdorff distance definition formulas, obtain:

H _MHD(A，B)＝max[h _MHD(A，B)，h _MHD(B，A)]

But average Hausdorff distance is relatively more responsive in the concentrated distribution of point to point.Be d by two parallel segments and Fig. 1 (b) by the average Hausdorff distance value between two point sets of two crossing line segment representatives as Fig. 1 (a), but two figure see have any different that from the angle of coupling the distribution of Fig. 1 (a) mid point should provide littler Hausdorff apart from the situation that just meets common people's perception than the distribution of Fig. 1 (b) mid point.This is its defective place.

Adopt the improved Hausdorff distance of standard variance and the intersegmental length difference of edge line then to overcome the defective of above-mentioned Hausdorff distance, by add the standard variance function and the edge length difference function of distance between point set on apart from the basis at average Hausdorff, introduced the distributed intelligence (Uniformity of Distribution between two point sets) of putting between point set and reached description, made it more accurate the difference between the edge line segment to length difference between the edge line segment.Simultaneously, final improvement Hausdorff distance has also reduced the influence that the outfield point is adjusted the distance and measured, and its adopts short line segment to remove to mate long line segment, does not need to ask maximin again.

So it is more careful by standard variance and the improved Hausdorff distance of the intersegmental length difference of edge line the difference between the edge line segment to be described, and has improved the discrimination that is used for people's edge image recognition in one's ear based on Hausdorff apart from the eigenwert that obtains.In using the final data sorting technique of support vector machine as ear recognition, adopt standard variance and the improved Hausdorff of the intersegmental length difference of edge line apart from extract the people in one's ear the edge proper vector discern, the discrimination of people's ear is greatly improved with respect to the ear recognition rate that adopts average Hausdorff distance and Hausdorff distance.

The present invention has overcome the Hausdorff distance noise-sensitive in the image and average Hausdorff distance is being put relatively sensitive issue of concentrated distribution to point.By pre-service to ear image, comprise denoising, bulk normalization and the illumination compensation etc. of the collection of ear image, non-colour of skin noise, obtain the standard ear image.Adopt again based on the method for gray scale morphology gradient and local Threshold Segmentation and extract people's edge feature in one's ear, obtain normal man's edge image in one's ear.On the basis of original Hausdorff distance, consider the concrete needs of in people's edge identification in one's ear, distinguishing marginal point and noise spot, propose a kind of based on standard variance and the improved Hausdorff of the intersegmental length difference of edge line apart from people's edge recognition methods in one's ear.The distributed intelligence that this method has been put between having added edge line segment point set in original Hausdorff distance and the description of the intersegmental length difference of edge line, it is more careful to make to normal man's portrayal of edge characteristics of image in one's ear, thereby has played the effect that improves the ear recognition rate.

Description of drawings

Fig. 1 is the defective of average Hausdorff distance, and wherein, Fig. 1 (a) is the integrated images of two parallel segment point sets, and Fig. 1 (b) is the integrated images of two crossing line segment point sets;

Fig. 2 is the algorithm flow chart of Face Detection;

Fig. 3 is that non-colour of skin noise is repaired algorithmic notation, wherein, and the two-dimensional matrix of Fig. 3 (a) presentation video brightness value, the subregion of the non-colour of skin noise of Fig. 3 (b) piece;

Fig. 4 is in order to calculate the chiasma type structural element b of ear image morphology gradient;

Fig. 5 is the normal man that obtains after handling edge image in one's ear;

Fig. 6 is 4 people edge line segment in one's ear that is used for computed improved Hausdorff distance.

Embodiment

Step 1: to collection and the pre-service and the ear image Edge Gradient Feature of ear image

(1) obtains the static colour picture that contains people's body side surface people ear by camera.The main color space of selecting for use the HSI space to cut apart as Face Detection, wherein, H (Hue) represents colourity, and S (Saturation) represents saturation degree, and I (Intensity) represents brightness.Adopt as Fig. 2 segmentation strategy, distinguish high saturation region of color and low saturation region, cut apart the high saturation region of color, cut apart the color harmony saturation region with I with H with S.After detecting non-colour of skin noise, adopt the reparation algorithm that the non-colour of skin noise region of ear image is repaired, concrete grammar is as follows:

A. the result who represents Face Detection with the matrix of a correspondence image size wherein has the gray-scale value 255 in the corresponding matrix of pixel of the colour of skin, but not the gray-scale value 0 in the corresponding matrix of the pixel of the colour of skin.Here gray-scale value is the notion of brightness, and 0-is a black, and 255-white is 0-255 according to the shade scope.Being in that the non-skin pixel piece of skin pixel in surrounding will be considered to is non-colour of skin noise region.The matrix of consequence of walkaway represents that with S the point that wherein has 0 value is a noise spot.

B. in a large amount of experiments, find, detected noise region always outline less than the noise region of reality.So, before adopting the reparation algorithm, enlarge noise region with conventional expansion algorithm:

SD＝dilate(S) (1)

Matrix of consequence after the SD representative is expanded in the formula (1), dilate () is known expansion function, S is the matrix of consequence of walkaway.

C. use the brightness value of matrix I presentation video.In matrix I, the pixel of each non-colour of skin noise region used one 5 * 5 matrix window, the window center point is processed non-skin pixel point (being with * number 0 value point among Fig. 3 (a)), skin pixel in window point (Fig. 3 (a) is middle with non-0 value representation) corresponding gray is sorted, and the pixel of choosing the intermediate grey values correspondence is a replacement pixel.The rgb value of using replacement pixel then is as the new rgb value of processed pixel.

D. shown in Fig. 3 (b) every non-colour of skin noise is equally divided into 4 districts by its height and width, each pixel in each zone all adopts the method in the C step to handle, and wherein, the A district adopts from top to bottom, processing sequence from left to right; The B district adopts from top to bottom, processing sequence from right to left; C district processing sequence is from top to bottom, from left to right; The processing sequence in D district then is from top to bottom, from right to left.The processing sequencing in each district is A, B, C, D.

E. after removing non-colour of skin noise, the space scale to ear image carries out conventional cutting and dwindles amplification again, carries out yardstick normalization, thereby obtains the calibration image of unified size.After finishing yardstick normalization, the image after the calibration is done gray scale normalization handle, do illumination compensation, improve the contrast of image with conventional figure image intensifying method histogram equalization.Obtain the standard ear image of 160*110 resolution at last.

(2) with structural element b input picture f is carried out gray scale and expand, be designated as With structural element b input picture f is carried out the gray scale corrosion, be designated as f Θ b.The morphology gradient of piece image is designated as g: $g=(f\⊕b)-\left(\mathrm{f\Θb}\right).$ Here adopt structural element b (as Fig. 4) to obtain the morphology gradient image of standard ear image.From these gradient images, be partitioned into edge of image.Adopt following segmentation strategy:

A. image is divided into the sub-piece that 100 sizes are 16 * 11 pixels.

B. make for morphology gradient image f that its gray average is M, if the gray-scale value of all pixels in certain sub-image all less than M/3, then this sub-image is not cut apart, assert does not wherein have the edge, and all pixels of this sub-image are put 0.

C. for other sub-image beyond the sub-image described in the B, at first use conventional maximum variance between clusters to obtain threshold value TH, cut apart this sub-image according to TH then.

Next, adopt the Refinement operation in the morphological image that edge image is carried out thinning processing, to obtain more careful edge image.Then, use the operation such as burr of cutting out and remove in the morphological image to remove the burr on the edge and the edge of segment.At last,, remove noise and pseudo-edge according to people's contour shape etc. of edge in one's ear, thus the people who obtains accurately describing people's helix exterior feature edge image in one's ear.By above-mentioned steps, the people that we will finally obtain edge image in one's ear is called " normal man is the edge image in one's ear ", as the foundation of follow-up identification.As Fig. 5 is " normal man is the edge image in one's ear " that the different ear images of 4 width of cloth obtain after treatment.

Step 2: the improvement of Hausdorff distance

(1) improves the Hausdorff distance with standard variance and the intersegmental length difference of edge line

Edge is by forming in one's ear for the people, and two different people coupling of edge in one's ear are two couplings between point set in a sense.Utilize Hausdorff apart from describing the point set similarity and being used widely in recent years by the method that feature point set mates.Given two finite point set A={a ₁, a ₂..., a _mAnd B={b ₁, b ₂..., b _n, the Hausdorff distance definition between them is as follows:

H(A，B)＝max[h(A，B)，h(B，A)] (2)

The improved Hausdorff of standard variance has been apart from having added the distributed intelligence of putting between point set, and the description of point set is become more accurately and stable.Yet the improved Hausdorff distance of standard variance still has the place of omission, because the length at edge also is an important feature.Need in the Hausdorff distance, add description to difference in length.As follows to its improvement of doing:

$h_{\mathrm{LSDMHD}}(A,B,k,t)=\frac{1}{N_A}\underset{a\∈A}{\mathrm{\Σ}}\underset{b\∈B}{\min }\left|\right|a-b\left|\right|+\mathrm{kS}(A,B)+\mathrm{t\ΔN}---\left(3\right)$

$h_{\mathrm{LSDMHD}}(B,A,k,t)=\frac{1}{N_B}\underset{b\∈B}{\mathrm{\Σ}}\underset{a\∈A}{\min }\left|\right|b-a\left|\right|+\mathrm{kS}(B,A)+\mathrm{t\ΔN}---\left(4\right)$

A in last two formulas, B represents finite point set A={a respectively ₁, a ₂..., a _mAnd B={b ₁, b ₂..., b _n.

N _A, N _BRepresent finite point set A respectively, the quantity of B mid point.

Δ N represents the difference of the quantity of putting between two point sets.

Parameter k is a weighting coefficient, is used for point of adjustment distributed intelligence shared proportion in distance calculation.

Parametric t is a weighting coefficient also, the difference in length that is used for regulating the edge line segment shared proportion in distance calculation.So just added description to difference in length between line segment.

S (A, B) standard variance of maximum distance in the point set B a bit among the expression point set A.

$S(A,B)=\sqrt{\underset{a\∈A}{\mathrm{\Σ}}{\left[\underset{b\∈B}{\min }\right||a-b||-\frac{1}{N_A}\underset{a\∈A}{\mathrm{\Σ}}\underset{b\∈B}{\min }||a-b|\left|\right]}^2}---\left(5\right)$

(B A) then represents among the point set B some standard variance of maximum distance in the point set A to S.

$S(B,A)=\sqrt{\underset{b\∈B}{\mathrm{\Σ}}{\left[\underset{a\∈A}{\min }\right||b-a||-\frac{1}{N_B}\underset{b\∈B}{\mathrm{\Σ}}\underset{a\∈A}{\min }||b-a|\left|\right]}^2}---\left(6\right)$

In last two formulas, A, B represent finite point set A={a respectively ₁, a ₂..., a _mAnd B={b ₁, b ₂..., b _n.

N _A, N _BRepresent finite point set A respectively, the quantity of B mid point.

Δ N represents the difference of the quantity of putting between two point sets:

ΔN＝||N _A-N _B|| (7)

N _A, N _BRepresent finite point set A respectively, the quantity of B mid point.

In order further to reduce the influence that the outfield point is adjusted the distance and measured, again the Hausdorff distance has been made following change, order:

$H_{\mathrm{LSDMHD}}(A,B,k,t)=\left\{\begin{array}{cc}{h}_{\mathrm{LSDMHD}}(A,B,k,t)& N_A\≤N_B\\ h_{\mathrm{LSDMHD}}(B,A,k,t)& N_A>N_B\end{array}\right.---\left(8\right)$

Alphabetical implication cotype (3) and formula (4) in the formula.

The Hausdorff distance that formula (8) provides is exactly the Hausdorff distance metric that finally utilizes after standard variance and the intersegmental length difference of edge line improve.

Step 3: adopt and improve Hausdorff distance determining to the human ear characteristic vector

As can be seen from Figure 6, the main edge contour of people's ear can roughly be divided into 4 sections, 4 edge line segment L that just indicated among the figure ₁, L ₂, L ₃, L ₄Practice shows by experiment, is feasible with these 4 line segments as 4 features of ear recognition.The collection A that sets up an office is the width of cloth " normal man is the edge image in one's ear " in the human ear characteristic storehouse, comprises 4 edge line segment L _A1, L _A2, L _A3, L _A4Point set B be a people to be identified in one's ear the edge image comprise 4 edge line segment L equally _B1, L _B2, L _B3, L _B4Now want test b whether to mate, can define the relative characteristic vector X of B and A with A _ABAs follows:

$X_{\mathrm{AB}}=\left[\begin{array}{c}{x}_1\\ x_2\\ x_3\\ x_4\end{array}\right]---\left(9\right)$

Adopt standard variance and the improved Hausdorff of the intersegmental length difference of edge line apart from extracting people's proper vector of edge image in one's ear:

x ₁＝H _LSDMHD(L _A1，L _B1) (10)

x ₂＝H _LSDMHD(L _A2，L _B2) (11)

x ₃＝H _LSDMHD(L _A3，L _B3) (12)

x ₄＝H _LSDMHD(L _A4，L _B4) (13)

Wherein, L _A1, L _A2, L _A3, L _A4, L _B1, L _B2, L _B3, L _B4Expression has the four edges edge line segment of ear image and ear image to be identified respectively.H _LSDMHDFor improving the Hausdorff distance metric.

Utilize existing recognition methods then,, obtain final recognition result as algorithm of support vector machine, neural network algorithm, K mean algorithm etc.What present embodiment adopted is algorithm of support vector machine.

With X _ABThe linear discriminant function d (X) of substitution point set A correspondence, in general

d(X _AB)＝W ⁱX _AB+W ₀ (14)

Here W ⁱBe sorter weight vector, W ₀But be setting parameter

If d is (X _AB)＞0 item shows that B and A are the edge of same people's ear, on the contrary, if d is (X _AB)≤0 an expression B and A are not the edges of same people's ear.

Implement 1

Gather 320 width of cloth people ear auris dextra images by digital camera, 320 width of cloth ear images in the ear image storehouse done following processing:

A. all ear images are carried out pretreatment operation, comprise denoising, space scale normalization and gray scale normalization.Everyone ear Flame Image Process is become " standard ear image ".

B. use edge detection method based on gray scale morphology gradient and local Threshold Segmentation, all " standard ear images " are done edge extracting, thereby obtain 320 width of cloth " normal man is the edge image in one's ear ", Here it is human ear characteristic storehouse, can use sparse matrix to describe " normal man is the edge image in one's ear ", to reduce data volume.

C. use the final data sorting technique of support vector machine as ear recognition.From the human ear characteristic storehouse, select 30 people's 150 width of cloth (everyone 5 width of cloth) " normal man is the edge image in one's ear " as training sample set; And remaining 90 width of cloth of this 30 people (everyone 3 width of cloth) " normal man is the edge image in one's ear " are as registered people's ear test sample book collection.All the other 10 people's 80 width of cloth (everyone 8 width of cloth) " normal man is the edge image in one's ear " are as unregistered people's ear test sample book collection.

D. adopt different Edge Distance measures to extract the edge line segment feature respectively, and use training sample set,, thereby obtain a complete human ear identification method for everyone trains a sorter.

The ear recognition rate that table 1 adopts three kinds of different Hausdorff distances to obtain

Test the discrimination of this method, obtain adopting the result of three kinds of different Hausdorff, see Table 1 apart from the ear recognition rate that obtains to people's ear.

From result of experiment we directly use as can be seen Hausdorff distance (Hausdorff Distance, HD), recognition effect is relatively poor, discrimination is also lower; Adopt average Hausdorff distance (ModifiedHausdorff Distance, MHD) the back discrimination all improves; Hausdorff distance (Length﹠amp after utilizing the intersegmental length difference of standard variance and edge line to improve; Standard Deviation ModifiedHausdorff Distance, LSDMHD) discrimination is the highest.

Conclusion: improved Hausdorff is apart from the discrimination height of ear recognition, and the discrimination of registered people's ear is reached 94.4%, and the reject rate of unregistered people's ear is reached 95.0%.。

Claims (4)

1. utilize and improve Hausdorff, it is characterized in that described method has following steps apart from the method for extracting the identification human ear characteristic:

(1) ear image of gathering is carried out pre-service:

By denoising and reparation to the non-colour of skin noise of the ear image gathered, adopt bulk normalization and illumination compensation, obtain the standard ear image;

(2) ear image Edge Gradient Feature:

Extract people's feature of edge in one's ear with the method that gray scale morphology gradient and local Threshold Segmentation combine, obtain normal man's edge image in one's ear;

(3) utilize the intersegmental length difference of standard variance and edge line to improve the Hausdorff distance, with the Hausdorff distance H after improving _LSDMHDBe the final Hausdorff distance metric that adopts after standard variance and the intersegmental length difference of edge line improve;

(4) with the people that obtained in the step (2) in one's ear its main edge contour of edge image be divided into 4 feature line segments of ear recognition, determine its human ear characteristic vector with improved Hausdorff distance.

2. utilization according to claim 1 improves Hausdorff apart from the method for extracting the identification human ear characteristic, it is characterized in that the step of the denoising of non-colour of skin noise in the step (1) and reparation is as follows:

The main color space of selecting for use the HSI space to cut apart as Face Detection, S distinguishes the high saturation region of color and low saturation region, H are cut apart the high saturation region of color and I is cut apart the color harmony saturation region, wherein the non-area of skin color that is surrounded by area of skin color is differentiated and is noise region, noise region by around area of skin color repair filling.

3. utilization according to claim 1 improves Hausdorff apart from the method for extracting the identification human ear characteristic, it is characterized in that the step of utilizing standard variance and the intersegmental length difference of edge line to improve the Hausdorff distance in the step (3) is as follows:

To difference in length with following formula to Hausdorff apart from description

$h_{\mathrm{LSDMHD}}(A,B,k,t)=\frac{1}{N_A}\underset{a\∈A}{\mathrm{\Σ}}\underset{b\∈B}{\min }\left|\right|a-b\left|\right|+\mathrm{kS}(A,B)+\mathrm{t\ΔN}$

$h_{\mathrm{LSDMHD}}(B,A,k,t)=\frac{1}{N_B}\underset{b\∈B}{\mathrm{\Σ}}\underset{a\∈A}{\min }\left|\right|b-a\left|\right|+\mathrm{kS}(B,A)+\mathrm{t\ΔN}$

Wherein, parameter k is used for the weighting coefficient of point of adjustment distributed intelligence proportion in distance calculation; Parametric t is the weighting coefficient that is used for regulating difference in length proportion in distance calculation of edge line segment; S (A, B) standard variance of maximum distance in the point set B a bit among the expression point set A,

$S(A,B)=\sqrt{\underset{a\∈A}{\mathrm{\Σ}}{\left[\underset{b\∈B}{\min }\right||a-b||-\frac{1}{N_A}\underset{a\∈A}{\mathrm{\Σ}}\underset{b\∈B}{\min }||a-b|\left|\right]}^2}$

S (B A) then represents among the point set B some standard variance of maximum distance in the point set A,

$S(B,A)=\sqrt{\underset{b\∈B}{\mathrm{\Σ}}{\left[\underset{a\∈A}{\min }\right||b-a||-\frac{1}{N_B}\underset{b\∈B}{\mathrm{\Σ}}\underset{a\∈A}{\min }||b-a|\left|\right]}^2}$

Δ N represents the difference of the quantity of putting between two point sets:

Δ N=||N _AOne N _B||

Wherein, N _A, N _BRepresent finite point set A respectively, the quantity of B mid point;

For reducing the influence that non-contour edge line segment point is adjusted the distance and measured, the Hausdorff distance has been made following change, order:

$H_{\mathrm{LSDMHD}}(A,B,k,t)=\left\{\begin{array}{cc}{h}_{\mathrm{LSDMHD}}(A,B,k,t)& N_A\≤N_B\\ h_{\mathrm{LSDMHD}}(B,A,k,t)& N_A>N_B\end{array}\right.$

H _LSDMHDBe the final Hausdorff distance metric that adopts after standard variance and the intersegmental length difference of edge line improve.

4. utilization according to claim 1 improves Hausdorff apart from the method for extracting the identification human ear characteristic, it is characterized in that improving in the described step (4) the Hausdorff distance and determines that the step of human ear characteristic vector is as follows:

The collection A that sets up an office is the width of cloth " normal man is the edge image in one's ear " in the human ear characteristic storehouse, comprises 4 edge line segment L _A1, L _A2, L _A3, L _A4, point set B is another people to be identified edge image in one's ear, comprises 4 edge line segment L _B1, L _B2, L _B3, L _B4, whether test b mates with A, the relative characteristic vector X of definition B and A _ABAs follows:

$X_{\mathrm{AB}}=\left[\begin{array}{c}{x}_1\\ x_2\\ x_3\\ x_4\end{array}\right]$

Wherein:

x ₁＝H _LSDMHD(L _A1，L _B1)

x ₂＝H _LSDMHD(L _A2，L _B2)

x ₃＝H _LSDMHD(L _A3，L _B3)

x ₄＝H _LSDMHD(L _A4，L _B4)。

Images