CN101059837A - Colorful contact lens false-proof detection method - Google Patents

Abstract

An anti-false check method of colorful contact lens comprises that 1, using geometry method to position the outer circle of iris, 2, based on the parameter of the outer circle, finding the constant area of a false iris, normalizing while the sample angle is 180-360degrees, 3, based on the normalized iris image in the step 2, extracting the contrast of a gray symbiotic matrix of the normalized iris image and the angle second-order moment, calculating the gray average value and variance of the normalized iris image, to obtain a four-dimension character vector, 4, based on the character vector, using a vector machine to classify two types, as one type is real iris and another type is false iris. The invention can quickly and effectively check colorful contact lens, to reduce the iris recognize error rate.

Description

The false-proof detection method of colorful contact lens

Technical field

What the present invention relates to is a kind of method of technical field of image processing, specifically is a kind of false-proof detection method of colorful contact lens.

Background technology

Based on one of authentication identifying method advantage of biological characteristic is to have higher reliability, but all kinds of biological characteristics also have the danger that is forged, no exception as the iris recognition that misclassification rate is minimum, also is faced with the danger of being attacked.Along with the application of iris recognition technology is more and more wider, the potential attack that existing iris authentication system suffers also can get more and more, present topmost pseudo-iris is to print the class iris, comprising: the living body iris image print that will illegally obtain is on the papery and be printed on the colorful contact lens.Therefore, domestic and international existing method for anti-counterfeit is primarily aimed at prints the pseudo-iris of class, and the false proof of pair colorful contact lens seldom arranged.

Find through literature search prior art, John Daugman is at " Journal of Wavelets, Multiresolution, and Information Processing " (small echo, multiresolution and information processing magazine) " Demodulation by complex-valuedwavelets for stochastic pattern recognition " (be used for random character identification based on multiple small echo demodulation) of delivering on (2003 the 1st phase 1-17 pages or leaves), the 14th page of spectral characteristic that has proposed to utilize the pseudo-iris of FFT change detection in this article, concrete grammar is: the iris image that will wear colorful contact lens earlier carries out 2 dimension FFT conversion, by detecting in the spectrogram whether 4 intermediate frequency spot zones are arranged, judge whether living body iris or pseudo-iris then.This method is utilized the print characteristic of printer itself, can produce periodic printing vestige, and being reflected on the frequency spectrum has very high energy on a certain frequency range, as long as therefore utilize the FFT conversion to detect.Its deficiency is: when the intentional out-focus of assailant, the method for then utilizing FFT to detect pseudo-iris will lose efficacy.

Summary of the invention

The objective of the invention is to overcome the deficiency of existing colorful contact lens method for anti-counterfeit on accuracy of detection, proposed a kind of false-proof detection method of colorful contact lens.It is false proof that the present invention utilizes the unchangeability of colorful contact lens texture region and gamma characteristic to carry out iris, and it can be detected existing colorful contact lens on the market, has real-time on speed.

The present invention is achieved by the following technical solutions, may further comprise the steps:

The first step is utilized method of geometry location iris cylindrical;

In second step, the parameter of the iris cylindrical that obtains according to the first step is determined the invariant region of pseudo-iris, carries out the normalization operation then, considers the interference of eyelashes and eyelid, and sampling angle is 180 to spend to 360 degree scopes during normalization;

In the 3rd step, the normalization iris image according to second step obtained carries out the feature extraction of pseudo-iris.Mainly extract contrast and two textural characteristics of angle second moment of the gray level co-occurrence matrixes of normalization iris image, calculate the gray average and the variance of normalization iris image simultaneously, obtain the proper vector of one 4 dimension;

In the 4th step,, carry out the identification of pseudo-iris according to the proper vector that the 3rd step obtained.Mainly utilize support vector machine to carry out the classification of two classes, that is, a class is true iris, and another kind of is pseudo-iris.

The described first step is meant: determine circle parameter in the iris earlier, determine the zone at cylindrical place then, utilize method of geometry to locate the iris cylindrical fast again.Specifically realize by following three steps:

1. determine circle parameter in the iris: according to the intensity profile characteristics of iris image, the gray-scale value of pupil is smaller and distribution is more even, and therefore available Gray Projection method is determined the central coordinate of circle of circle in the iris;

2. the edge extraction of iris cylindrical: at first, central coordinate of circle according to circle in the iris of determining, (iris radius that collects has individual approximate range in conjunction with priori, occurrence depends on different collecting devices) can determine iris cylindrical place approximate range, then it is dwindled 30% ratio, utilize the Canny edge detection method that the iris region that dwindles is carried out edge extraction.Utilize Canny to carry out rim detection, detected edge noise is few and be single pixel edge, during detection, detect operator and be adjusted to vertical direction, because in vertical direction, the outward flange of iris is blocked less by eyelashes and eyelid, then can comprise a lot of noises on the horizontal direction, the parameter of circle is removed noise informations such as pupil region and eyelid in the iris of utilize determining, further removes random noise with eight connectivity criteria more at last, obtains removing the outline map behind the noise.

3. the location of iris cylindrical: the location that the edge image after utilizing method of geometry to denoising is justified.If the center of circle of pupil is for a some P, in outline map, it is capable to scan n along some P downwards with size delta level uniformly-spaced, and these sweep traces and outline map are crossing, get outer peripheral focus, obtain two groups of point set B and C, wherein, and B={B _i| i=1,2,3 ... n}, C={C _i| i=1,2,3 ... n}.Then can utilize not any 3 calculation of parameter that the principle that can determine its circumscribed circle is justified point-blank, promptly from two groups of point set B and C, randomly draw at 3 and calculate radius of a circle r _iAnd central coordinate of circle (x _i, y _i), wherein, i=1,2,3 ... m, m are the number of times of randomly drawing.Therefore the radius and the central coordinate of circle that obtain this iris cylindrical are respectively: $r=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{r}_i$ With $(x=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i,y=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{y}_i).$ At last according to scale down, by the original iris cylindrical of the coaptation parameter of the circle of downscaled images.

In described second step, be meant: the invariant region of determining pseudo-iris according to the parameter of iris cylindrical.Promptly get outer region away from pupil, because the texture of pseudo-iris can not change along with illumination (can cause the contraction of pupil), therefore can estimate the roughly constant scope of pseudo-iris, consider the interference of eyelashes and eyelid, during normalization, the actual samples angle is 180 to spend to 360 degree scopes.

Described the 3rd step, be meant: can define such as features such as: contrast, angle second moment (claim not only energy), local homogeney (but also claiming unfavourable balance value square) and correlativitys based on gray level co-occurrence matrixes, wherein, contrast con can be understood as sharpness, i.e. the readability of texture.The texture rill is dark more in the image, and contrast is big more, and visual effect is clear more.For open grain, the con value is less, and for close grain, the con value is bigger; Angle second moment asm is a kind of tolerance to the image texture distributing homogeneity, i.e. the right repeatability of pixel.When gradation of image distributed relatively evenly, the asm value was bigger, otherwise the asm value is then less, and angle second moment and first-order statistics amount (such as: contrast and variance) uncorrelated fully, promptly the value when contrast and variance is that 0 hour angle second moment may reach maximal value.

The present invention has utilized contrast con and two features of angle second moment asm of gray level co-occurrence matrixes, in addition, the gradation of image statistical nature, that is: also as the feature of colorful contact lens, these 4 characterizing definitions are as follows for the gray average m of normalization iris image and variances sigma:

$m=$ $\frac{1}{W\×H}\underset{x=1}{\overset{H}{\mathrm{\Σ}}}\underset{y=1}{\overset{W}{\mathrm{\Σ}}}I(x,y)---\left(1\right)$

$\mathrm{\σ}=\sqrt{\frac{1}{W\×H}\underset{x=1}{\overset{H}{\mathrm{\Σ}}}\underset{y=1}{\overset{W}{\mathrm{\Σ}}}{(I(x,y)-m)}^2}---\left(2\right)$

$\mathrm{con}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{(i-j)}^{2}P(i,j)---\left(3\right)$

$\mathrm{asm}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}P{(i,j)}^2---\left(4\right)$

Wherein, contrast (con) can be understood as sharpness, i.e. the readability of texture.The texture rill is dark more in the image, and contrast is big more, and visual effect is clear more.For open grain, the con value is less, and for close grain, the con value is bigger; Angle second moment (asm) is a kind of tolerance to the image texture distributing homogeneity, i.e. the right repeatability of pixel.When gradation of image distributed relatively evenly, the asm value was bigger, otherwise the asm value is then less.

Therefore, a width of cloth colorful contact lens image can be characterized by the proper vector of one 4 dimension, and this proper vector is defined as follows:

V＝[m，σ，con，asm] ^T (5)

In described the 4th step, be meant: utilize support vector machine to classify, that is, a class is true iris, and another kind of is pseudo-iris, and support vector machine is used RBF (Gauss is base nuclear radially) function.

$K(x,x_i)=\exp \{-\frac{{|x-x_i|}^2}{{\mathrm{\σ}}^2}\}---\left(6\right)$

Wherein, x is for treating classification samples, x _iBe support vector, σ is the standard variance of RBF.

The present invention carries out pseudo-iris feature and extracts on the basis of iris cylindrical location, because pseudo-iris texture gray scale is darker, and gray scale is more even, and true iris can shrink along with the variation pupil of mistake photograph and the texture gray scale is more shallow, therefore extract contrast and two textural characteristics of angle second moment of the gray level co-occurrence matrixes of normalization iris image, calculate the gray average and the variance of normalization iris image simultaneously, amount to 4 proper vectors, utilize support vector machine to carry out the classification of two classes at last.

Compared with prior art, the iris method for anti-counterfeit of the present invention's proposition has higher precision and practicality.(include the true iris image of 2000 width of cloth with the iris storehouse of gathering, 250 width of cloth are worn the iris image of colorful contact lens) test, wherein, wear the iris image of colorful contact lens trains with the true iris image of 1000 width of cloth and 150 width of cloth, all the other are used for testing, experimental situation is Matlab 7.1, WindowsXP, Pentium 4 3GHz512M internal memories.Test result is: discrimination is 100%.On speed, identification of the present invention is 252.8 milliseconds (wherein, 190.5 milliseconds consuming time of Iris Location, 16.2 milliseconds consuming time of normalization is extracted 31.3 milliseconds consuming time of eigenwert, 14.8 milliseconds consuming time of pseudo-iris recognition) averaging time.Experiment shows that the colorful contact lens detection method that the present invention proposes can both satisfy real-time requirement on speed and precision.

Description of drawings

Fig. 1 is embodiment of the invention cylindrical positioning result figure.

The pseudo-iris invariant region result schematic diagram that Fig. 2 extracts for the embodiment of the invention.

Fig. 3 is an embodiment of the invention normalization result schematic diagram.

Embodiment

Below in conjunction with accompanying drawing embodiments of the invention are elaborated: present embodiment has provided detailed embodiment and process being to implement under the prerequisite with the technical solution of the present invention, but protection scope of the present invention is not limited to following embodiment.

In order to verify validity of the present invention, the iris image that embodiment uses adopts the iris storehouse of oneself taking, and wherein colorful contact lens is the common color contact lenses of selling on the market.Whole implement process is as follows:

(1) utilize method of geometry to locate the iris cylindrical fast: earlier, to determine the zone at cylindrical place then, utilize method of geometry to locate the iris cylindrical fast again.Specifically realize by following three steps:

1. determine circle parameter in the iris: according to the intensity profile characteristics of iris image, the gray-scale value of pupil is smaller and distribution is more even, and therefore available Gray Projection method is determined the central coordinate of circle of circle in the iris;

2. the edge extraction of iris cylindrical: at first, according to the central coordinate of circle of determining circle in the iris, in conjunction with priori (present embodiment, the iris maximum radius is estimated as 150 pixels) can determine iris cylindrical place approximate range, then it is dwindled certain ratio (present embodiment uses 30%), as Fig. 1. (a).Utilize the Canny edge detection method that the iris region that dwindles is carried out edge extraction, during rim detection, detect operator and be adjusted to vertical direction, as Fig. 1. (b).The parameter of circle is removed noise informations such as pupil region and eyelid in the iris of utilize estimating, is communicated with further removal random noise with eight more at last, obtains removing the outline map behind the noise, as Fig. 1. (c).

3. the location of iris cylindrical: the location that the edge image after utilizing method of geometry to denoising is justified.If the center of circle of pupil is a some P, coordinate be P (x, y)=P (35,39), in outline map, P scans n (getting 4) downwards OK with size delta (getting 6) level uniformly-spaced along point, and these sweep traces and outline map are crossing, get outer peripheral focus, obtain two groups of point set B and C, wherein, B={B _i| i=1,2,3 ... n}, C={C _i| i=1,2,3 ... n} is as Fig. 1. (d).Then can utilize not any 3 calculation of parameter that the principle that can determine its circumscribed circle is justified point-blank, promptly from two groups of point set B and C, randomly draw at 3 and calculate radius of a circle r _iAnd central coordinate of circle (x _i, y _i), wherein, i=1,2,3 ... m, the number of times of m (getting 10) for randomly drawing.Therefore the radius and the central coordinate of circle that obtain this iris cylindrical are respectively: $r=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{r}_i$ With $(x=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i,y=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{y}_i).$ At last according to scale down, by the original iris cylindrical of the coaptation parameter of the circle of downscaled images.This cylindrical center of circle is [row, col]=[230,361], and radius is 136, and the cylindrical positioning result as shown in Figure 2.

(2) determine the invariant region of pseudo-iris according to the parameter of iris cylindrical, promptly get outer region away from pupil, this zone radius is made as 70 pixels (original iris image size is 640 * 480), because the texture of pseudo-iris can not change along with illumination (can cause the contraction of pupil), therefore can estimate the roughly constant scope of pseudo-iris, consider the interference of eyelashes and eyelid, the actual samples angle is 180 to spend to 360 degree scopes (as shown in Figure 3) during normalization, and the normalization result as shown in Figure 4.

(3) pseudo-iris feature extracts, and extracts contrast and two textural characteristics of angle second moment of the gray level co-occurrence matrixes of normalization iris image, calculates the gray average and the variance of normalization iris image simultaneously, amount to 4 statistics texture feature vectors, the proper vector V=[m of gained, σ, con, asm] ^T=[77.44,3.34,0.1366,0.2644] ^T

(4) pseudo-iris recognition: utilize support vector machine to carry out the classification of two classes, that is, a class is true iris, another kind of is pseudo-iris, support vector machine is used RBF (Gauss is base nuclear radially) function, and the Gauss radially parameter of base nuclear is made as: standard deviation is 3, and the coboundary is 10.

Classification results in the result images can come out the pseudo-iris detection of iris well as can be seen, has higher precision, and simultaneously, the living body iris method for anti-counterfeit that present embodiment proposes has higher speed.Above embodiment is always consuming time to be 252.4 milliseconds (wherein, 190.2 milliseconds consuming time of Iris Location, 16.2 milliseconds consuming time of normalization, 31.3 milliseconds consuming time of extraction eigenwert, 14.7 milliseconds consuming time of pseudo-iris recognition).

Claims (4)

1, a kind of false-proof detection method of colorful contact lens is characterized in that, may further comprise the steps:

The first step utilizes method of geometry to determine cylindrical location iris cylindrical: to determine circle parameter in the iris earlier, determine the zone at cylindrical place then, utilize method of geometry promptly to locate the iris cylindrical according to 3 definite cylindricals again;

In second step, the parameter of the iris cylindrical that obtains according to the first step is determined the invariant region of pseudo-iris, carries out the normalization operation then, and sampling angle is 180 to spend to 360 degree scopes during normalization;

The 3rd step, go on foot the normalization iris image that obtains according to second, carry out the feature extraction of pseudo-iris, extract contrast and two textural characteristics of angle second moment of the gray level co-occurrence matrixes of normalization iris image, calculate the gray average and the variance of normalization iris image simultaneously, obtain the proper vector of one 4 dimension;

The 4th step, according to the proper vector that the 3rd step obtained, carry out the identification of pseudo-iris, utilize support vector machine to carry out the classification of two classes, that is, a class is true iris, another kind of is pseudo-iris.

2, the false-proof detection method of colorful contact lens according to claim 1 is characterized in that, the described first step, by following three the step realize:

1. determine the central coordinate of circle of circle in the iris with the Gray Projection method;

2. the edge extraction of iris cylindrical: at first, according to circle parameter in the iris of 1. determining, determine iris cylindrical place approximate range in conjunction with priori, then it is dwindled 30% ratio, utilize the Canny edge detection method that the iris region that dwindles is carried out edge extraction, during detection, detect operator and be adjusted to vertical direction, the parameter of circle is removed pupil region and eyelid noise information in the iris of utilize estimating, is communicated with further removal random noise with eight more at last, obtains removing the outline map behind the noise;

3. the location of iris cylindrical: the location that the edge image after utilizing method of geometry to denoising is justified, the center of circle of promptly establishing pupil is a P, in outline map, it is capable to scan n along some P downwards with size delta level uniformly-spaced, these sweep traces and outline map intersect, and get outer peripheral focus, obtain two groups of point set B and C, wherein, B={B _i| i=1,2,3...n}, C={C _i| i=1,2,3...n}; Utilize not any 3 calculation of parameter that the principle of determining its circumscribed circles is justified point-blank, promptly from two groups of point set B and C, randomly draw at 3 and calculate radius of a circle r _iAnd central coordinate of circle (x _i, y _i), wherein, i=1,2,3...m, m are the number of times of randomly drawing, the radius and the central coordinate of circle that therefore obtain this iris cylindrical are respectively: $r=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{r}_i$ With $(x=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i,y=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{y}_i),$ At last according to scale down, by the original iris cylindrical of the coaptation parameter of the circle of downscaled images.

3, the false-proof detection method of colorful contact lens according to claim 1, it is characterized in that, in described the 3rd step, adopt the contrast con and the angle second moment asm of gray level co-occurrence matrixes, and the gray average m of normalization iris image and variances sigma be as the feature of colorful contact lens, and these 4 characterizing definitions are as follows:

$m=\frac{1}{W\×H}\underset{x=1}{\overset{H}{\mathrm{\Σ}}}\underset{y=1}{\overset{W}{\mathrm{\Σ}}}I(x,y)$

$\mathrm{\σ}=\sqrt{\frac{1}{W\×H}\underset{x=1}{\overset{H}{\mathrm{\Σ}}}\underset{y=1}{\overset{W}{\mathrm{\Σ}}}{(I(x,y)-m)}^2}$

$\mathrm{con}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{(i-j)}^{2}P(i,j)$

$\mathrm{asm}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}P{(i,j)}^2$

One width of cloth colorful contact lens characterization image is the proper vector of one 4 dimension, and this proper vector is defined as follows:

V＝[m，σ，con，asm] ^T。

4, the false-proof detection method of colorful contact lens according to claim 1 is characterized in that, in described the 4th step, support vector machine is used the radially basic kernel function of Gauss,

$K(x,x_i)=\exp \{-\frac{{|x-x_i|}^2}{{\mathrm{\σ}}^2}\}.$

Wherein, x is for treating classification samples, x _iBe support vector, σ is the radially standard variance of base nuclear of Gauss.

Images