CN102081739B - Iris characteristic extracting method based on FIR (Finite Impulse Response) filter and downsampling - Google Patents

Abstract

The invention provides an iris characteristic extracting method based on FIR (Finite Impulse Response) filter and downsampling, which comprises the following steps of: 1, carrying out one-dimensional treatment on an acquired two-dimensional iris image; 2, filtering a one-dimensional iris signal through the FIR filter, wherein the obtained filtered one-dimensional iris signal is used as an extracted iris characteristic; and 3, carrying out downsampling and characteristic coding on the filtered characteristic, determining a downsampling factor by using the highest cut-off frequency of the FIR fitler, downsampling the filtered iris characteristic signal according to the downsampling factor to obtain the downsampled iris characteristic signal, and carrying out Hamming coding on the downsampled iris characteristic signal, wherein a characteristic vector subjected to the Hamming coding is an extracted iris characteristic vector. The characteristic in the iris, which is most beneficial to the identification, is extracted by using the filtering idea; and the iris characteristic extracting method has the characteristics of simpleness of realizing, less quantity of identification characteristics and high identification speed.

Description

Iris feature method for distilling based on FIR wave filter and down-sampling

Technical field

The present invention relates to a kind of iris feature method for distilling that is used for iris recognition, belong to iris feature extractive technique field.

Background technology

Iris recognition technology is a kind of important biometrics identification technology.Compare with the other biological recognition technology, it has high stability, and high antifalsification does not have advantages such as the property of infringement and high accuracy, receives increasing concern, and vast potential for future development is arranged.

Typical iris recognition comprises several committed steps such as iris image acquiring, iris image pre-service, feature extraction and coupling identification.Iris image acquiring general using infrared light supply and infrared lens are realized.The iris that not only comprises similar toroidal in the iris image that collects also has pupil, sclera, eyelid, eyelashes even the reflective hot spot that causes etc., need carry out corresponding pre-service to reduce the influence that these factors are brought.Iris image obtains normalized iris image through after the pre-service, and texture wherein contains number of characteristics information, can be used for identifying a people's identity.The task that iris feature extracts is exactly effectively to extract these characteristic informations, and it is described as being convenient to compare the proper vector or the feature coding of classification.Through feature extraction; Iris image is replaced by corresponding proper vector or feature coding; The work that coupling identification will be done is exactly the similarity of comparing between iris image proper vector or the feature coding, and judges according to certain rule whether two width of cloth irises belong to same individual.

Feature extraction is the core of iris recognition technology, is directly connected to the effect of identification.Current, the classical approach method that iris feature extracts mainly contains: the 2-D Gabor wave filter that utilizes that Daugman proposes carries out filtering to iris, and the phase information of extracting iris texture is as characteristic; Wilds utilizes the laplacian pyramid two dimensional filter that iris is decomposed, and utilizes the texture image under the different resolution to discern; Boles and Boashash calculates the small echo zero crossing of iris texture and representes to discern; L.Ma utilizes the local strength of iris to change, and utilizes Gaussian-Hermite square or small echo to carry out feature extraction.In addition, some feature extraction schemes that also have other.But the number of features that these methods are extracted is more, causes taking more storage space, and has reduced recognition speed.

Summary of the invention

The present invention is directed to existing iris feature method for distilling; A kind of iris feature method for distilling based on FIR wave filter and down-sampling simple, that the recognition feature number is few, recognition speed is fast of realizing is provided, and this method utilizes filtering thought to extract the characteristic that helps discerning most in the iris.

Iris feature method for distilling based on FIR wave filter and down-sampling of the present invention, to object be one dimension iris signal, specifically may further comprise the steps:

(1) generation of one dimension iris signal

The iris two dimensional image of gathering is carried out of one-dimensional to be handled; Through the iris two dimensional image that collects is carried out Iris Location, normalization and histogram equalization pretreatment operation; Obtain pretreated normalization rectangular image; Last 50% 50% iris region of pupil (promptly near) zone of choosing the normalization rectangular image connects the area-of-interest part as area-of-interest by row, obtain one dimension iris signal.

The object that the present invention is directed to is an one dimension iris signal, utilizes one dimension FIR wave filter (limit for length's unit impulse response filter is arranged) to realize feature extraction.Because the iris image that utilizes iris capturing device (iris image acquiring general using infrared light supply and infrared lens are realized) to obtain is a two dimensional image, at first iris image is carried out of one-dimensional and handle.At first to the iris image that collects position, pretreatment operation such as normalization, histogram equalization, obtain pretreated normalization rectangular image, its size is K * L (K representes the width of rectangular image, and L representes the length of rectangular image).Owing in the iris image that collects existence such as eyelid, eyelashes are often arranged, can influence the extraction of iris feature.For fear of the influence that eyelid, eyelashes cause blocking of iris, last 50% 50% iris region of pupil (promptly near) that the present invention chooses normalized image as area-of-interest, carries out of one-dimensional and handles.

$I=\left(\begin{array}{c}{I}_1\\ .\\ .\\ .\\ I_i\\ .\\ .\\ .\\ I_K\end{array}\right)={(I_1^T,...,T_i^T,...I_K^T)}^T(i=\mathrm{1,2},......,K)---\left(1\right)$

Following formula has provided the expression formula of two-dimentional iris image.Wherein, I _iRepresent the capable gray-scale value of i in the normalized image.

50% iris region interested near pupil is carried out of one-dimensional handles, mainly be iris image with two dimension through converting one-dimensional signal into by the row ways of connecting, its connected mode is following:

V＝{I ₁，I ₂，…，I _k，…I _K/2}＝{v ₁，v ₂，…，v _j，…v _n}

Wherein, V is the one dimension iris signal after connecting by row, and what mainly get here is that preceding K/2 is capable, I _iThe capable gray-scale value of i of presentation video I, v _jBe illustrated in the gray-scale value that position j is ordered in the one dimension vector V, the dimension of the one-dimensional vector that n representes to be constituted.The present invention is primarily aimed at V and carries out feature extraction.

(2) extract the iris feature signal based on one dimension FIR wave filter

One dimension iris signal is carried out filtering through the FIR wave filter, and as the iris feature signal that extracts, the cutoff frequency of FIR wave filter, the logical scope of band and exponent number are selected according to the separability criterion with the filtered one dimension iris signal that obtains.

The FIR filter table is shown fir, and the logical scope of its cutoff frequency and band is: [F _Start-F _End], filter order is N, three parameter values here confirm that main passing through realized according to the separability criterion the experiment of sample storehouse:

$J_d=\frac{\mathrm{tr}\left(S_b\right)}{\mathrm{tr}\left(S_w\right)}---\left(2\right)$

Wherein, S _bBe dispersion between class, S _wBe dispersion in the class, J _dA dispersion and type ratio of interior dispersion between type of being defined as, its value is big more, and then classifying quality is good more.

Make J _dThe maximum filter parameter of value be optimal parameter, at this moment, can extract the best frequency band range in the best frequency band feature feature set of iris, utilize wave filter fir that one dimension iris signal V is carried out filtering:

W＝V*Fir (3)

Wherein, W={w ₁, w ₂..., w _j... W _nBe filtered iris signal, * representes convolution, has so just realized the extraction of iris signal characteristic.

Utilize the FIR wave filter to carry out feature extraction and mainly contain following advantage:

(i) parameter can the flexible setting, to the validity of iris recognition, and selects best filter parameter to discern under the convenient checking different parameters.And parameter is in case confirm that need not make amendment, realization is simple.

(ii) have the linear phase characteristics, can guarantee after filtering, to keep the phase information of iris texture.

(iii) compare with iir filter, stability is higher, more helps the application under the large sample situation.

(3) down-sampling of characteristic is handled and feature coding after the filtering

Higher cutoff frequency by the FIR wave filter is confirmed the down-sampling factor, filtered iris feature signal is carried out down-sampling according to the down-sampling factor handle, and obtains the iris feature signal behind the down-sampling; Iris feature signal to behind the down-sampling carries out hamming code, and the iris feature vector behind the hamming code is the iris feature vector that extracts.

Generally speaking, the filtered iris signal data of FIR amount equates that with original iris signal data amount data volume is bigger, need carry out down-sampling to filtered iris signal characteristic and handle.The logical scope of filtered iris signal band is [F _Start-F _End], its highest SF is F _EndAccording to nyquist sampling theorem, if with 2F _EndSF filtered iris signal is carried out down-sampling, can realize reconstruct to the iris signal.Therefore, the present invention carries out down-sampling to filtered iris signal W and handles, and W is carried out D extraction down-sampling doubly, the down-sampling factor D:

D＝floor[F _s/(2F _end)] (4)

Wherein, F _sBe the normalization SF, floor representes that the ratio result rounds downwards, and the value of D is the ratio of normalization SF and the higher cutoff frequency twice of wave filter.Utilize D to carry out the signal that down-sampling obtains and satisfy sampling thheorem, guaranteed the integrality of signal characteristic.With the iris signal W ' behind the D times of down-sampling as proper vector.At this moment, the length of full feature vector has only the 1/D of original iris signal, has reduced the data volume of characteristic greatly, can more effective realization to the storage of iris database, save the space storage resources; Simultaneously, improved the speed of coupling identification.

For the characteristic vector W that obtains ', also need carry out feature coding, realize the further compression of storage data and the further raising of recognition speed.Here, the present invention adopts the hamming code scheme:

${w^{\&prime;}}_j=\left\{\begin{array}{cc}1,& \mathrm{if}{w}_j\&GreaterEqual;0\\ 0,& \mathrm{if}{w}_j<0\end{array}\right.---\left(5\right)$

So just realized coding to signal characteristic.W ' behind the coding is and finally is used to the proper vector discerned, the iris feature that is just extracted.

The present invention utilizes filtering thought to extract the characteristic that helps discerning most in the iris, has the characteristics simple, that the recognition feature number is few, recognition speed is fast that realize.

Embodiment

Iris feature method for distilling based on FIR wave filter and down-sampling of the present invention specifically may further comprise the steps:

(1) utilizes existing general image harvester (current iris image gathers the general using infrared light supply and infrared lens is realized) to realize collection, obtain the iris two dimensional image iris image.Iris two dimensional image to gathering carries out pre-service such as Iris Location, normalization, histogram equalization, obtains rectangle normalized image I, and size is K * L.

(2) choose iris feature zone and generate one dimension iris signal.Choose the area-of-interest of last 50% zone of normalization iris image I as feature extraction, at this moment, the area-of-interest size obtains one dimension iris signal for K/2 * L. connects the area-of-interest part by row:

V＝{I ₁，I ₂，…，I _k，…I _K/2}＝{v ₁，v ₂，…，v _j，…v _n}

At this moment, one dimension iris signal V length is n=K * L/2.

(3) designing optimal FIR feature extraction wave filter.Utilize the separability criterion of formula (2), the logical scope of design band is [F _Start-F _End], exponent number is the FIR wave filter (fir) of N, is used to extract the best frequency band feature of iris.

(4) one-dimensional signal filtering feature extraction.V carries out filtering through fir with one dimension iris signal:

W＝V*Fir

Obtain filtered iris signal W={w ₁, w ₂..., w _j... W _n, as the iris feature that extracts.

(5) down-sampling of iris feature signal is handled.According to formula (4), confirm the down-sampling factor D by the higher cutoff frequency of wave filter, with filtered iris feature signal W, carry out down-sampling according to the down-sampling factor D and handle, obtain the characteristic signal W '={ w behind the down-sampling ₁, w ₂..., w _j... W _n, at this moment, the signal length of W ' is the 1/D. of W

(6) feature coding.Iris signal W ' to behind the down-sampling carries out hamming code:

${w^{\&prime;}}_j=\left\{\begin{array}{cc}1,& \mathrm{if}{w}_j\&GreaterEqual;0\\ 0,& \mathrm{if}{w}_j<0\end{array}\right.$

Characteristic vector W behind the hamming code ', be the proper vector that extracts.

By above step (1)--the extraction of iris feature can be realized in (6), the characteristic of extraction can be used for the storage and the identification of iris signal.

Claims (1)

1. the iris feature method for distilling based on FIR wave filter and down-sampling is characterized in that, may further comprise the steps:

(1) generate one dimension iris signal:

The iris two dimensional image of gathering is carried out of one-dimensional to be handled; Through the iris two dimensional image that collects is carried out Iris Location, normalization and histogram equalization pretreatment operation; Obtain pretreated normalization rectangular image; Last 50% zone of choosing the normalization rectangular image connects the area-of-interest part as area-of-interest by row, obtain one dimension iris signal;

(2) extract the iris feature signal based on one dimension FIR wave filter:

One dimension iris signal is carried out filtering through the FIR wave filter, and as the iris feature signal that extracts, the cutoff frequency of FIR wave filter, the logical scope of band and exponent number are selected according to the separability criterion with the filtered one dimension iris signal that obtains; The separability criterion does

In the formula, S _bBe dispersion matrix between the class of sample, S _wBe the within class scatter matrix of sample, J _dThe ratio of the mark of dispersion matrix trace and within class scatter matrix between type of being defined as;

(3) down-sampling of characteristic is handled and feature coding after the filtering:

Higher cutoff frequency by the FIR wave filter is confirmed the down-sampling factor, filtered iris feature signal is carried out down-sampling according to the down-sampling factor handle, and obtains the iris feature signal behind the down-sampling; Iris feature signal to behind the down-sampling carries out hamming code, and the iris feature vector behind the hamming code is the iris feature vector that extracts.