CN101241550A - Iris image quality judging method - Google Patents

Abstract

The present invention provides a method of iris image quality judgement, firstly normalizing operation to extract original normalizing iris image, then processing four-layer multi-resolution decomposition of the normalizing iris image, the image whether existing eyelid and lash shading problem or not is judged by counting number of the larger amplitude value of the details weight under the resolving capability 2-1, and comparing with the predetermined threshold value, then calculating the variance of the details weight under the resolving capability 2-2, and comparing with the predetermined threshold value can judge whether the image definition meets the requirement of system or not. The method of iris image quality judgement of the invention has stable ability to the different lighting environment, and quickly and exactly judges the iris image quality.

Description

A kind of iris image quality judging method

[technical field]

The invention belongs to technical field of image processing, relate generally to the iris identity recognizing technology in the biological characteristic discriminating.

Background technology]

Along with electronic equipments such as computer, ATM, mobile phone, access control system constantly enter in our daily life, for the personal security, identity identifying technology becomes more and more urgent easily.Yet existing system based on smart card, ID (identity number) card No. and password but can only pace up and down in safety and conveniently, and safety never realized fully, and better safety occurs simultaneously with inconvenience.In order to realize higher security, we must use more complicated and more inconvenient password, because if our different at one's side machine is used an identical password, we have also increased the hidden danger of security when having obtained convenience.People are exploring some safety and solutions easily always for this reason.Biological identification technology provides a brand-new field for this reason.Biological identification technology is to utilize human physiological property or behavioral trait to carry out identification and checking.These physiological properties comprise fingerprint, palmmprint, and sound, signature waits the identity of discerning the individual.Because physiological property neither can pass into silence as password, can not lost, so be considered to a kind of more reliable personal identification verification method as key yet.The iris identification is a kind of emerging biological identification technology, utilizes the foundation of iris as identification, has high uniqueness, high stability, natural antifalsification and does not have advantage such as infringement property.See document for details: Anil K.Jain, Arun Ross, SalilPrabhakar, " An Introduction to Biometric Recognit ion ", IEEETransaction on Circuits and Systemsfor Video Technology, Volume 14, No.1, pp.4-20,2004 and document: John G.Daugman, " How Iris RecognitionWorks; " IEEE Transaction on Circuits and Systems for Video Technology, Volume 14, and Issue 1, pp.21-30,2004 is described.

It is a very important part that iris image quality is evaluated in the whole automatic iris recognition technology, and it has guaranteed that the iris image of handling all satisfies the quality requirements of system.Thereby, avoided the mistake knowledge that causes owing to the quality problems of iris image own and refused knowledge.In the reality, because the focal length problem of collecting device when taking, the rotational problems of moment eyeball, and eyelid and eyelashes usually make the iris image of collection can't carry out follow-up feature extraction to the partial occlusion of iris.In the existed algorithms a kind of effective iris image quality assessment models is not proposed also at present, therefore we are intended to set up the general feasible assessment models of a cover, see document for details: Chen Ji, Hu Guangshu, " Iris Image Quality Evaluation based onWavelet Packet Decomposition ", Journal of Tsinghua University (Sci ﹠amp; Tech), Volume 43, No.3, pp.377-380,2003 and document: Li Ma, Tieniu Tan, Yunhong Wang, Dexin Zhang, " Efficient Iris Recognition byCharacterizing key Local Variations ", IEEE Transaction on ImageProcessing, Volume 13, No.6, pp.739-750,2004 is described.

At present existing iris quality determination methods has:

(1) based on the method for fast fourier transform.It carries out fast two-dimensional fourier transformation to the picture element in two rectangular blocks on the iris region, and then by to the statistics of its high frequency, intermediate frequency and low frequency energy, whether analysis image is clear and exist eyelashes to block.The versatility of this model is not strong, and easily that texture is less clear iris image erroneous judgement is the inferior quality iris image.See document for details: Li Ma, Tieniu Tan, YunhongWang, Dexin Zhang, " Personal Identification based on Iris TextureAnalysis, " IEEE Transactions on Patern Analysis and MachineIntelligence, Volume 25, No.12, pp.1519-1533.

(2) based on the method for WAVELET PACKET DECOMPOSITION.It is chosen the texture high fdrequency component and distributes the most concentrated sub-band as the feature sub-band, with the criterion of its energy as differentiation picture quality.The shortcoming of this method is can't judge because of eyelashes to block in-problem iris image.See document for details: Chen Ji, Hu Guangshu, " Iris Image Quality Evaluation based on Wavelet Packet Decomposition ", Journal of Tsinghua University (Sci﹠amp; Tech), Volume 43, No.3, pp.377-380,2003.

(3) based on the method for image definition, inside and outside degree of eccentricity and iris visibility.It has set up image definition, inside and outside degree of eccentricity and three indexs of weighing picture quality of iris visibility, has realized iris image is carried out the requirement of real-time quality assessment.The shortcoming of this method is comparatively responsive to illumination condition, and stability is not strong.Xing?Lei，Shi?Pengfei，“A?Quality?Evaluation?Method?of?Iris?Image”，Chinese?Journal?of?Stereology?and?Image?Analysis，Volume.8，No.2，pp.108-113，2003。

Above-mentioned iris image quality evaluation algorithm is existing problems to a certain extent all, and calculated amount is excessive,, versatility comparatively responsive to illumination is strong or the like.

[summary of the invention]

The objective of the invention is to set up the more intense iris image quality judging method of a kind of versatility, can detect iris image and the not enough iris image of sharpness that the eyelid eyelashes block accurately, and the illumination condition scope that algorithm is suitable for is wider.

The object of the present invention is achieved like this;

A kind of iris image quality judging method is characterized in that it comprises the following step:

Step 1, by camera head, the iris in the human eye is carried out image acquisition, from the original-gray image that contains iris image, obtain being of a size of M * N normalization iris image f (x, y).(x, the y) coordinate of remarked pixel point, f (x, y) denotation coordination is (x, gray values of pixel points y);

Step 2, the normalization iris image to obtaining in the step 1 carry out 4 layers of two-dimensional wavelet transformation; Specifically, the formula of two-dimensional wavelet transformation is: With $W_{\mathrm{\ψ}}^t(j,m,n)=\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}f(x,y){\mathrm{\ψ}}_{j,m,n}^t(x,y);$ Wherein, W _(j, m, n) and W _ψ ⁱ(j, m are respectively that resolution is 2 n) ^jUnder scale coefficient and wavelet coefficient, the span of 4 layers of wavelet transformation j be 1 ,-2 ,-3 ,-4}, i={H, V, D} has added level, vertical and to the details of angular direction;  _{J, m, n}(x y) is scaling function, ψ _{J, m, n} ⁱ(x y) is wavelet function, and the small echo of choosing is the DMeyer small echo;

Step 3, by the horizontal direction wavelet coefficient W under the resolution 2-1 that obtains in the step 2 _ψ ^H(1, m, n), vertical direction wavelet coefficient W _ψ ^V(1, m is n) with to angular direction wavelet coefficient W _ψ ^D(1, m, n) the original normalization iris image of reconstruct is in resolution 2 ^-1Under details component D _2-1, (x y), that is to say high frequency details component to f; Specifically, reconstruction formula is: $D_{2^j}f(x,y)=$ $\underset{i=\{H,V,D\}}{\mathrm{\Σ}}\underset{m}{\mathrm{\Σ}}\underset{n}{\mathrm{\Σ}}{W}_{\mathrm{\ψ}}^t(j,m,n){\mathrm{\ψ}}_{j,m,n}^t(x,y);$ D wherein _2j, f (x, y) expression resolution 2 ^jFollowing coordinate (x, the value of details component y), W _ψ ⁱ(j, m, n) expression resolution 2 ^jUnder wavelet coefficient, ψ _{J, m, n} ⁱ(x y) is the 2-d wavelet of i orientation-sensitive.∑ is the accumulating operation symbol;

The high frequency details component D that obtains in step 4, the calculation procedure 3 _2-1F (x, absolute value y) | D _2-1F (x, y) |.Specifically, if $D_{2^{-1}}f(x,y)\&GreaterEqual;0,$ $\left|D_{2^{-1}}f(x,y)\right|=D_{2^{-1}}f(x,y);$ If $D_{2^{-1}}f(x,y)<0$ $\left|D_{2^{-1}}f(x,y)\right|=-D_{2^{-1}}f(x,y);$

Obtain in step 5, the statistic procedure 4 | D _2-1(x, y) | value is greater than V for f _oThe number of pixel, will | D _2-1(x, y) | value is greater than V for f _oPoint as the frontier point of eyelid and eyelashes; Concrete computing formula is:

Wherein M and N are respectively the width and the height of original normalization iris image, $n(x,y)=\left\{\begin{array}{cc}1& \left|D_{2^{-1}}I(x,y)\right|\&GreaterEqual;V_o\\ 0& \left|D_{2^{-1}}I(x,y)\right|<V_o\end{array};\right.$

Calculate in step 6, the comparison step 5

Value and threshold value T _NMake comparisons.If

Think that then this image is the image that eyelid and eyelashes block, if

Think that then this image is accessible normal iris image;

Step 7, according to step 3 in similar method, by the resolution 2 that obtains in the step 2 ^-2Under horizontal direction wavelet coefficient W _ψ ^H(2, m, n), vertical direction wavelet coefficient W _ψ ^V(2, m is n) with to angular direction wavelet coefficient W _ψ ^D(2, m, n) the original normalization iris image of reconstruct is in resolution 2 ^-2Under details component D _2-2F (x, y);

The D that obtains in step 8, the calculation procedure 7 _2-2F (x, variance y); Concrete computing formula is: $\mathrm{Var}=\left[\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}{(D_{2^{-2}}I(x,y)-m)}^2\right]/M\&times;N;$ Wherein, Var is a resolution 2 ^-2The variance of following details component, M and N are respectively the width and the height of original normalization iris image, and the computing formula of m is: $m=\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}{D}_{2^{-2}}I(x,y)/M\&times;N;$

Step 9, with the variance Var that obtains in the step 8 and the threshold value T that differentiates the iris sharpness _VCompare, if Var 〉=T _V, then this iris image sharpness satisfies system requirements; If Var＜T _V, the then discontented pedal system requirement of this iris image sharpness.

Aforesaid a kind of iris image quality judging method is characterized in that carrying out in the step 2 the scaling function  of two-dimensional wavelet transformation _{J, m, n}(x y) is two unidimensional scale function  _{J, m}(x) and  _{J, n}(y) product.Carry out the wavelet function ψ of the horizontal direction sensitivity of two-dimensional wavelet transformation _{J, m, n} ^H(x y) is one dimension wavelet function ψ _{J, m}(x) and unidimensional scale function  _{J, n}(y) product; Carry out the wavelet function ψ of the vertical direction sensitivity of two-dimensional wavelet transformation _{J, m, n} ^V(x y) is unidimensional scale function  _{J, m}(x) and one dimension wavelet function ψ _{J, n}(y) product; Carry out two-dimensional wavelet transformation and drink the wavelet function ψ of diagonal angle orientation-sensitive _{J, m, n} ^D(x y) is two one dimension wavelet function ψ _{J, m}(x) and ψ _{J, n}(y) product.

Aforesaid a kind of iris image quality judging method is characterized in that the threshold value T that mentions in the step 6 _NBe to be used to judge whether this image is the image that manageable no eyelid and eyelashes block, T _NDetermine relevant with follow-up matching algorithm, be set into Iris image coupling, the T of correspondence when misclassification rate is minimum _N

The present invention has adopted the method for multiresolution analysis, by analyzing the details component under the different resolution, correct evaluation the quality of iris image.The present invention at first positions original iris image, obtains the normalization iris image.By the normalization iris image being carried out 4 layers of wavelet transformation, statistics resolution 2 ^-1Under the details component big amplitude point number and compare with pre-set threshold, judge whether this iris image exists eyelid and eyelashes occlusion issue.Afterwards, statistics resolution 2 ^-2Under the variance of details component, and compare with pre-set threshold, judge whether iris image clear.The analysis that utilizes the thought of multiresolution to carry out iris image quality is a characteristic of the present invention, compares with general iris image quality appraisal procedure, and versatility of the present invention and stability are very strong, are not subject to the influence of illumination.

[description of drawings]

Fig. 1 is the original image that contains iris;

Wherein, 1 expression pupil; 2 expression irises; Hot spot in the 3 expression pupils; The inner edge of 4 expression irises; The outer rim of 5 expression irises.

Fig. 2 is positioning result figure and normalization figure;

Wherein, (a) be positioning result figure; (b) be the iris normalized image.

Fig. 3 is original normalization iris image and high frequency details component thereof;

Wherein, be the normalization iris image that has eyelid to block (a), (b) be the high frequency details component of (a), be the normalization iris image that ciliation blocks (c), (d) be the high frequency details component of (c).

[embodiment]

In order to describe content of the present invention easily, at first some terms are defined.

Definition 1: iris.The center of eyeball is the pupil of black, and the outer intermarginal annular tissue of pupil is iris.It presents the textural characteristics of interlaced similar and spot, filament, striped, crypts.Same individual's iris can change in life hardly the people's, and the iris of different people is different fully.

Definition 2: gray level image.Only comprise monochrome information in the image and without any the image of other colouring informations.

Definition 3: normalization iris image.After original iris image positioned, the head rotation when taking for eliminating, shooting distance far and near inconsistent, the image that problem such as pupil convergent-divergent and carrying out obtains after normalization is operated, the normalization iris image has identical size.

Definition 4: wavelet transformation.The localization analytical approach of time (space) frequency, it progressively carries out multiple dimensioned refinement by flexible translation computing to signal (function), can focus on any details of signal.

Definition 5; Scale coefficient.In carrying out wavelet transformation, original signal and scaling function carry out the coefficient that convolution obtains afterwards, are used for the approximate component of reconstruction signal.For two-dimensional wavelet transformation, the concrete computing formula of scale coefficient is:

Wherein, (x y) is original signal,  to f _{J, m, n}(x y) is scaling function, W _(j, m n) are scale coefficient.

Definition 6: scaling function.The expanded function set that scaling function is made up of integer translation and real number two-value yardstick, quadractically integrable function  (x), i.e. set {  _{J, m}(x) }. wherein _{J, m}(x)=2 ^J/2 (2 ^jX-m), j, m ∈ Z.The scaling function  of the two-dimensional wavelet transformation in the definition 5 _{J, m, n}(x y) is two unidimensional scale function  _{J, m}(x) and  _{J, n}(y) product.

Definition 7: wavelet coefficient.When carrying out wavelet transformation, original signal and wavelet function carry out the coefficient that convolution obtains afterwards, are used for the details component of reconstruction signal.For two-dimensional wavelet transformation, the concrete computing formula of wavelet coefficient is: $W_{\mathrm{\&psi;}}^t(j,m,n)=\underset{x=0}{\overset{M-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}f(x,y){\mathrm{\&psi;}}_{j,m,n}^t(x,y).$ Wherein, (x y) is original signal, ψ to f _{J, m, n} ^j(x y) is wavelet function, W _ψ ⁱ(j, m n) are wavelet coefficient.

Definition 8: wavelet function.Wavelet function is to be used for describing the difference of crossing over adjacent two metric spaces, by the expanded function set that ψ (x) forms, promptly gathers { ψ _{J, k}(x) }.ψ wherein _{J, m}(x)=2 ^J/2ψ (2 ^jX-m), j, m ∈ Z.The wavelet function ψ of the horizontal direction sensitivity of the two-dimensional wavelet transformation in the definition 5 _{J, m, n} ^H(x y) is one dimension wavelet function ψ _{J, m}(x) and unidimensional scale function  _{J, n}(y) product; Carry out the wavelet function ψ of the vertical direction sensitivity of 2-d wavelet exchange _{J, m, n} ^V(x y) is unidimensional scale function  _{J, m}(x) and one dimension wavelet function ψ _{J, n}(y) product; Carry out the wavelet function ψ of the diagonal angle orientation-sensitive of two-dimensional wavelet transformation _{J, m, n} ^D(x y) is two one dimension wavelet function ψ _{J, m}(x) and ψ _{J, n}(y) product.

Definition 9:DMeyer small echo.The Meyer small echo of discrete form is the effectively approximate of Meyer small echo, can be regarded as the Meyer small echo of discretize, has biorthogonality.It had both kept the good frequency division characteristic of Meyer small echo, can improve the speed of numerical evaluation again.

Definition 10: multiresolution analysis.The thought of multiresolution analysis is meant that mainly (x, y) regarding resolution as is being similar under 2 °=1, this approximate coarse resolution 2 that further is decomposed into original image f ^JApproximate component under (J＜0) and a series of high resolution 2 ^j(the progressive sum of approaching of details component under the j＞J).

Definition 11: details component.Any piece image can be decomposed into main information and detail textures information, and according to the thought of multiresolution analysis, the details component refers to the detail textures information in the different frequency range scope.

Definition 12: coupling.A certain concrete things correctly is included into a certain classification.

Definition 13.Misclassification rate.Of a sort things is included into the probability of other classifications, and FMR commonly used represents.

According to iris image quality judging method of the present invention, it comprises the following step:

Step 1, by camera head, the iris in the human eye is carried out image acquisition, from the original-gray image that contains iris image, obtain being of a size of M * N normalization iris image f (x, y).(x, the y) coordinate of remarked pixel point, f (x, y) denotation coordination is (x, gray values of pixel points y);

Step 2, the normalization iris image to obtaining in the step 1 carry out 4 layers of two-dimensional wavelet transformation; Specifically, the formula of two-dimensional wavelet transformation is:

With $W_{\mathrm{\&psi;}}^i(j,m,n)=\underset{x=0}{\overset{M-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}f(x,y){\mathrm{\&psi;}}_{j,m,n}^i(x,y);$ Wherein, W _(j, m, n) and W _ψ ⁱ(j, m are respectively that resolution is 2 n) ^jUnder scale coefficient and wavelet coefficient, the span of 4 layers of wavelet transformation j be 1 ,-2 ,-3 ,-4}, i={H, V, D} has added level, vertical and to the details of angular direction;  _{J, m, n}(x y) is scaling function, ψ _{J, m, n} ⁱ(x y) is wavelet function, and the small echo of choosing is the DMeyer small echo;

Step 3, resolution 2 by obtaining in the step 2 ^-1Under horizontal direction wavelet coefficient W _ ^H(1, m, n), vertical direction wavelet coefficient W _ ^V(1, m is n) with to angular direction wavelet coefficient W _ψ ^D(1, m, n) the original normalization iris image of reconstruct is in resolution 2 ^-1Under details component D _2-1(x y), that is to say high frequency details component to f; Specifically, reconstruction formula is: $D_{2^i}f(x,y)=$ $\underset{i=\{H,V,D\}}{\mathrm{\&Sigma;}}\underset{m}{\mathrm{\&Sigma;}}\underset{n}{\mathrm{\&Sigma;}}{W}_{\mathrm{\&psi;}}^i(j,m,n){\mathrm{\&psi;}}_{j,m,n}^i(x,y);$ D wherein _2jF (x, y) expression resolution 2 ^jFollowing coordinate (x, the value of details component y), W _ψ ⁱ(j, m, n) expression resolution 2 ^jUnder wavelet coefficient, ψ _{J, m, n} ^j(x y) is the 2-d wavelet of i orientation-sensitive.∑ is the accumulating operation symbol;

The high frequency details component D that obtains in step 4, the calculation procedure 3 _2-1F (x, absolute value y) | D _2-1F (x, y) |, specifically, if $D_{2^{-1}}f(x,y)\&GreaterEqual;0,$ $\left|D_{2^{-1}}f(x,y)\right|=D_{2^{-1}}f(x,y);$ If $D_{2^{-1}}f(x,y)<0,$ $\left|D_{2^{-1}}f(x,y)\right|={-D}_{2^{-1}}f(x,y);$

Obtain in step 5, the statistic procedure 4 | D _2-1(x, y) | value is greater than V for f _oThe number of pixel, will | D _2-1(x, y) | value is greater than V for f _oPoint as the frontier point of eyelid and eyelashes; Concrete computing formula is: Wherein M and N are respectively the width and the height of original normalization iris image, $n(x,y)=\left\{\begin{array}{cc}1& \left|D_{2^{-1}}I(x,y)\right|\&GreaterEqual;V_o\\ 0& \left|D_{2^{-1}}I(x,y)\right|<V_o\end{array};\right.$

Calculate in step 6, the comparison step 5

Value and threshold value T _NMake comparisons.If

Think that then this image is the image that eyelid and eyelashes block, if

Think that then this image is accessible normal iris image:

Step 7, according to step 3 in similar method, by the resolution 2 that obtains in the step 2 ^-2Under horizontal direction wavelet coefficient W _ψ ^H(2, m, n), vertical direction wavelet coefficient W _ψ ^V(2, m is n) with to angular direction wavelet coefficient W _ψ ^D(2, m, n) the original normalization iris image of reconstruct is in resolution 2 ^-2Under details component D _2-2F (x, y);

The D that obtains in step 8, the calculation procedure 7 _2-2F (x, variance y); Concrete computing formula is: $\mathrm{Var}=\left[\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}{(D_{2^{-2}}I(x,y)-m)}^2\right]/M\&times;N;$ Wherein, Var is a resolution 2 ^-2The variance of following details component, M and N are respectively the width and the height of original normalization iris image, and the computing formula of m is: $m=\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}{D}_{2^{-2}}I(x,y)/M\&times;N;$

Step 9, with the variance Var that obtains in the step 8 and the threshold value T that differentiates the iris sharpness _VCompare, if Var 〉=T _V, then this iris image sharpness satisfies system requirements; If Var＜T _V, the then discontented pedal system requirement of this iris image sharpness.

By above step, we just can extract normalized iris image by analysis and judge the requirement whether this image satisfies system from the original image that contains iris.

Need to prove:

1. the operation of the iris normalization in the step 1 must be to carry out after Iris Location.

2. carry out the scaling function  of two-dimensional wavelet transformation in the step 2 _{J, m, n}(x y) is two unidimensional scale function  _{J, m}(x) and  _{J, n}(y) product.Carry out the wavelet function ψ of the horizontal direction sensitivity of two-dimensional wavelet transformation _{J, m, n} ^H(x y) is one dimension wavelet function ψ _{J, m}(x) and unidimensional scale function  _{J, n}(y) product; Carry out the wavelet function ψ of the vertical direction sensitivity of two-dimensional wavelet transformation _{J, m, n} ^V(x y) is unidimensional scale function  _{J, m}(x) and one dimension wavelet function ψ _{J, n}(y) product; Carry out the wavelet function ψ of the diagonal angle orientation-sensitive of two-dimensional wavelet transformation _{J, m, n} ^D(x y) is two one dimension wavelet function ψ _{J, m}(x) and ψ _{J, n}(y) product.

3. step 3 is selected resolution 2 in 4,5 ^-1Under the details component carry out the analysis that eyelid and eyelashes block, be because at the edge of eyelid and eyelashes, have tangible grey scale change on the gray level image, according to the thought of multiresolution analysis, the some correspondence of these positions the big amplitude of high frequency details component.

4. the threshold value V in the step 5 _oBe used to judge whether the point on the high frequency details component is the marginal point of eyelid and eyelashes.Think high frequency details component amplitude | D _2-1F (x, y) | greater than V _oPoint be exactly the marginal point of eyelid and eyelashes, be because at the marginal position place of eyelid and eyelashes, have the obvious saltus step of gray-scale value, the bigger point of amplitude on corresponding the high frequency details component; And the texture marginal position place of iris, greyscale transformation is comparatively slow, the less point of amplitude on corresponding the high frequency details component.When high frequency details component amplitude | D _2-1F (x, y) | during greater than a certain value, this point is exactly the marginal point of eyelid and eyelashes.

5. the threshold value T that mentions in the step 6 _NBe to be used to judge whether this image is the image that manageable no eyelid and eyelashes block.T _NDetermine relevant with follow-up matching algorithm, we be set into

Iris image coupling, the T of correspondence when misclassification rate is minimum _N

6. step 7 is chosen the normalization iris image in resolution 2 in 8 ^-2Under details component D _2-2(x y) carries out the analysis of iris sharpness to f, is that according to the thought of multiresolution analysis, this variation is embodied in intermediate frequency details component D more because iris texture changes comparatively slowly on gray level image _2-2F (x, y).Iris image is clear more, and the variance Var of intermediate frequency details component is big more; Iris image is fuzzy more, and the variance Var of intermediate frequency details component is more little.

7. the threshold value T in the step 9 _VBe whether clearly to be used to judge iris image, T _VDetermine relevant with follow-up matching algorithm.

The present invention adopts the method for multiresolution analysis, at first by extracting the iris in the original-gray image and carrying out normalization; Then the normalization iris image is carried out multiresolution and decompose, obtain resolution 2 respectively ^-1With resolution 2 ^-2Under the details component; At last according to resolution 2 ^-1Under the number of big amplitude point of details component judge whether this image exists eyelid and eyelashes occlusion issue, according to resolution 2 ^-2The variance of following details component judges whether iris image is clear.The method based on multiresolution analysis that adopts the present invention to propose can effectively be carried out the evaluation of iris image quality, has avoided traditional algorithm to illumination sensitive issue comparatively.

Adopt method of the present invention, at first use C language compilation iris image quality appraisal procedure; Adopt the original image of CMOS or CCD camera head automatic shooting iris then; Then the iris original image that photographs is input in the iris image quality appraisal procedure on the PC platform as source data and handles; Comment through Iris Location, normalization and picture quality and to provide the judgement whether image satisfies system requirements after robust and sturdy.Adopt 2400 to take different illumination conditions good, that comprise different people, the different gray scale iris image of posture of taking as source data, the result of program judgement and the result of subjective judgement are compared, error probability is 1.2%, every width of cloth treatment of picture time＜150ms.

In sum, method of the present invention makes full use of the texture information of iris, in conjunction with the method for multiresolution analysis, thereby has realized judging fast and accurately the quality of iris image.

Claims (3)

1, a kind of iris image quality judging method is characterized in that it comprises the following step:

Step 1, by camera head, the iris in the human eye is carried out image acquisition, from the original-gray image that contains iris image, obtain being of a size of M * N normalization iris image f (x, y).(x, the y) coordinate of remarked pixel point, f (x, y) denotation coordination is (x, gray values of pixel points y);

Step 2, the normalization iris image to obtaining in the step 1 carry out 4 layers of two-dimensional wavelet transformation; Specifically, the formula of two-dimensional wavelet transformation is:

With $W_{\mathrm{\&psi;}}^t(j,m,n)=\underset{x=0}{\overset{M-1}{\mathrm{\&Sigma;}}}\underset{y=0}{\overset{N-1}{\mathrm{\&Sigma;}}}f(x,y){\mathrm{\&psi;}}_{j,m,n}^t(x,y);$ Wherein, W _(j, m, n) and W _ψ ⁱ(j, m are respectively that resolution is 2 n) ^jUnder scale coefficient and wavelet coefficient, the span of 4 layers of wavelet transformation j be 1 ,-2 ,-3 ,-4}, i={H, V, D} has added level, vertical and to the details of angular direction;  _{J, m, n}(x y) is scaling function, ψ _{J, m, n} ⁱ(x y) is wavelet function, and the small echo of choosing is the DMeyer small echo;

Step 3, by the horizontal direction wavelet coefficient W under the resolution 2-1 that obtains in the step 2 _ψ ^H(1, m, n), vertical direction wavelet coefficient W _ψ ^V(1, m is n) with to angular direction wavelet coefficient W _ψ ^D(1, m, n) the original normalization iris image of reconstruct is in resolution 2 ^-1Under details component D _2-1(x y), that is to say high frequency details component to f; Specifically, reconstruction formula is: $D_{2^j}f(x,y)=$ $\underset{i=\{H,V,D\}}{\mathrm{\&Sigma;}}\underset{m}{\mathrm{\&Sigma;}}\underset{n}{\mathrm{\&Sigma;}}{W}_{\mathrm{\&psi;}}^i(j,m,n){\mathrm{\&psi;}}_{j,m,n}^i(x,y);$ D wherein _2jF (x, y) expression resolution 2 ^jFollowing coordinate (x, the value of details component y), W _ψ ⁱ(j, m, n) expression resolution 2 ^jUnder wavelet coefficient, ψ _{J, m, n} ⁱ(x y) is the 2-d wavelet of i orientation-sensitive.∑ is the accumulating operation symbol;

The high frequency details component D that obtains in step 4, the calculation procedure 3 _2-1F (x, absolute value y) | D _2-1F (x, y) |.

Specifically, if $D_{2^{-1}}f(x,y)\&GreaterEqual;0,$ $\left|D_{2^{-1}}f(x,y)\right|=-D_{2^{-1}}f(x,y);$ If $D_{2^{-1}}f(x,y)<0,$ $\left|D_{2^{-1}}f(x,y)\right|=-D_{2^{-1}}f(x,y);$

Obtain in step 5, the statistic procedure 4 | D _2-1(x, y) | value is greater than V for f _oThe number of pixel, will | D _2-1(x, y) | value is greater than V for f _oPoint as the frontier point of eyelid and eyelashes; Concrete computing formula is: Wherein M and N are respectively the width and the height of original normalization iris image, $n(x,y)=\left\{\begin{array}{cc}1& \left|D_{2^{-1}}I(x,y)\right|\&GreaterEqual;V_o\\ 0& \left|D_{2^{-2}}I(x,y)\right|<V_o\end{array},;\right.$

Calculate in step 6, the comparison step 5

Value and threshold value T _NMake comparisons.If

Think that then this image is the image that eyelid and eyelashes block, if Think that then this image is accessible normal iris image;

Step 7, according to step 3 in similar method, by the resolution 2 that obtains in the step 2 ^-2Under horizontal direction wavelet coefficient W _ψ ^H(2, m, n), vertical direction wavelet coefficient W _ψ ^V(2, m is n) with to angular direction wavelet coefficient W _ψ ^D(2, m, n) the original normalization iris image of reconstruct is in resolution 2 ^-2Under details component D _2-2F (x, y);

The D that obtains in step 8, the calculation procedure 7 _2-zF (x, variance y); Concrete computing formula is: $\mathrm{Var}=\left[\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}{(D_{2^{-2}}I(x,y)-m)}^2\right]/M\&times;N;$ Wherein, Var is a resolution 2 ^-2The variance of following details component, M and N are respectively the width and the height of original normalization iris image, and the computing formula of m is: $m=\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}{D}_{2^{-2}}I(x,y)/M\&times;N;$

Step 9, with the variance Var that obtains in the step 8 and the threshold value T that differentiates the iris sharpness _VCompare, if Var 〉=T _V, then this iris image sharpness satisfies system requirements; If Var＜T _V, the then discontented pedal system requirement of this iris image sharpness.

2, a kind of iris image quality judging method according to claim 1 is characterized in that carrying out in the step 2 the scaling function  of two-dimensional wavelet transformation _{J, m, n}(x y) is two unidimensional scale function  _{J, m}(x) and  _{J, n}(y) product.Carry out the wavelet function ψ of the horizontal direction sensitivity of two-dimensional wavelet transformation _{J, m, n} ^H(x y) is one dimension wavelet function ψ _{J, m}(x) and unidimensional scale function  _{J, n}(y) product; Carry out the wavelet function ψ of the vertical direction sensitivity of two-dimensional wavelet transformation _{J, m, n} ^V(x y) is unidimensional scale function  _{J, m}(x) and one dimension wavelet function ψ _{J, n}(y) product; Carry out the wavelet function ψ of the diagonal angle orientation-sensitive of two-dimensional wavelet transformation _{J, m, n} ^D(x y) is two one dimension wavelet function ψ _{J, m}(x) and ψ _{J, n}(y) product.

3, a kind of iris image quality judging method according to claim 1 is characterized in that the Fujian value T that mentions in the step 6 _NBe to be used to judge whether this image is the image that manageable no eyelid and eyelashes block, T _NDetermine relevant with follow-up matching algorithm, be set into

Iris image coupling, the T of correspondence when misclassification rate is minimum _N