虹膜识别算法的纹理分析

Iris Recognition Algorithms Based on Texture AnalysisRichard Yew Fatt NgComputer Vision and Intelligent Systems (CVIS)Group Universiti Tunku AbdulRahman, Malaysia richng01@Yong Haur TayComputer Vision andIntelligent Systems (CVIS)GroupUniversiti Tunku AbdulRahman, Malaysiatayyh@.myKai Ming MokComputer Vision andIntelligent Systems (CVIS)GroupUniversiti Tunku AbdulRahman, Malaysiamokkm@.myAbstractIris recognition has become a popular research in recent years due to its reliability and nearly perfect recognition rates. Iris recognition system has three main stages: image preprocessing, feature extraction and template matching. An innovative method is proposed to extract iris features based on texture analysis. Iris textures are analyzed to capture the discriminating frequency information. Specific filters with different center frequency are applied to three different zones to extract the texture of the iris. Different weightings are given to each zone depending on its contribution to the recognition. The encoded binary templates are compact in size and can avoid the visibility of the individual iris images. The templates are suitable for implementing iris recognition devices using DSP (Digital Signal Processor). The proposed method was evaluated using CASIA iris image database version 1.0 [1]. Experimental results show that the proposed approach has achieved high accuracy of 98.62%.1. IntroductionBiometric identification is an emerging technology which gains more attention in recent years. Iris has distinct phase information which spans about 249 degrees of freedom [2,3]. This advantage let iris recognition be the most accurate and reliable biometric identification.The three main stages of an iris recognition system are image preprocessing, feature extraction and template matching.The iris image needs to be preprocessed to obtain useful iris region. Image preprocessing is divided into three steps: iris localization, iris normalization and image enhancement. Iris localization detects the inner and outer boundaries of iris. Eyelids and eyelashes that may cover the iris region are detected and removed. Iris normalization converts iris image from Cartesian coordinates to Polar coordinates. The iris image has low contrast and non-uniform illumination caused by the position of the light source. All these factors can be compensated by applying local histogram equalization.The paper proposed an innovative method for feature extraction and template matching stages. The iris region is divided into three zones according to the characteristic of the iris texture. Texture of the iris in each zone is analyzed in terms of the discriminating frequency information. Log Gabor filters with different center frequency are chosen accordingly to extract iris most significant texture features. Different weightings are selected for each zone based on its contribution to the recognition. The encoded binary templates are compact in size and can prevent the visibility of individual iris images. The templates can be stored and processed effectively using DSP technology.Iris segmentation method is presented in next section. Section 3 discusses iris normalization and image enhancement algorithms. Section 4 and Section 5 describe details of the proposed feature extraction and template matching stages. Experimental results and discussions are illustrated in Section 6 while the conclusion is drawn in the last section.2. Iris segmentationThis section discusses the iris segmentation method. It includes iris inner and outer boundaries localization, upper and lower eyelids detection andeyelashes, reflection and pupil noise removal algorithms.2.1. Iris inner boundary localizationAs pupil is a black circular region, it is easy to detect the pupil inside an eye image. Firstly, pupil is detected using thresholding operation. An appropriate threshold is selected to generate the binary image which contains pupil only. Morphological operator is applied to the binary image to remove the reflection inside the pupil region and other dark spots caused by eyelashes.Since the inner boundary of an iris can be approximately modeled as circles, circular Hough transform is used to localize the iris [4]. Firstly, edge detector is applied to binary image to generate the edge map. Gaussian filter is applied to smooth the image to select the proper scale of edge analysis. The center coordinate and radius of the circle with maximum number of edge points are defined as the pupil center and iris inner boundary respectively.2.2. Iris outer boundary localizationIn order to locate the iris outer boundary, the proposed method selects two search regions including the outer iris boundary. The pupil center is referred as origin. The search region is a sector with radius from pupil boundary to a maximum radius.The intensities of each radius in the two search regions are added up. The right and left iris boundary is the maximum difference between the sum of intensities of two outer radius and two inner radius.(a) (b)Figure 1.(a) Right and left search regions ofthe iris image. (b) Iris inner and outerboundaries localization.2.3. Upper and lower eyelids detectionTwo search regions are selected to detect upperand lower eyelids. The pupil center, iris inner and outerboundaries are used as reference to select the twosearch regions.(a) (b)Figure 2. (a) Upper and lower search regionsof the iris image. (b) Upper and lower eyelidsdetection.Sobel edge detection is applied to the searchregions to detect the eyelids. In order to reduce thefalse edges detection caused by eyelashes, Sobel kernelis tuned to the horizontal direction.After edge detection step, the edge image isgenerated. The eyelids are detected using linear HoughTransform method.2.4. Eyelashes, reflection and pupil noiseremovalThe eyelashes and pupil noises are observed tohave lower intensity values. A simple thresholdingtechnique is applied to segment eyelashes and pupilnoises accurately.Reflection regions are characterized by highintensity values close to 255. A high threshold valuecan be used to separate the reflection noise.Figure 3. Normalized iris image with pupil,eyelashes and reflection noises.3. Normalization and enhancementIris may be captured in different size with varyingimaging distance. Due to illumination variations, theradial size of the pupil may change accordingly.Therefore, the iris region needs to be normalized tocompensate for these variations. Figure 3 shows the irisimage after normalization.Normalization remaps each pixel in the localizediris region from the Cartesian coordinates to polarcoordinates. The non-concentric polar representation isnormalized to a fixed size rectangular block.Figure 4. Normalization process.)](,[θn p R R r ∈ ]2,0[πθ∈)cos(θ×+=r x x c i (1) )sin(θ×+=r y y c i (2)where (x i ,y i ) denotes the polar coordinates of a point inside iris region. (x c ,y c ) and R p are the center coordinates and radius of the pupil respectively. R n (θ) is the distance from pupil center to the iris outer boundary which is in the function of θ. It can be calculated using Equation (3) and (4).Figure 5. Geometry representation for irisnormalization.(x d ,y d ) and R s denote center coordinates and radius of the iris respectively.22)()(c d c d y y x x d −+−= (3) θθθcos )sin ()(22×+×−=d d R R s n (4)The normalized iris image has low contrast and non-uniform illumination caused by the light source position. The image needs to be enhanced to compensate for these factors.Local histogram equalization is applied to the normalized iris image to reduce the effect of non-uniform illumination and obtain well-distributed texture image.Figure 6. Enhanced iris image.Figure 7. (a) Three zones of the iris image. (b) Pupillary zone, collarette boundary andciliary zone of the iris.The iris region is divided into three zones according to the characteristic of the iris texture. Zone Z1, Z2 and Z3 are illustrated in Figure 6 and 7. Iris is divided into two major regions: pupillary zone and ciliary zone. Zone Z1 is the pupillary zone which contains abundant texture information. Collarette boundary is the part of iris separating pupillary zone and ciliary zone. It often appears in the middle zone Z2. Zone 3 corresponds to the ciliary zone with flat textures.4. Feature extractionIris has abundant of unique texture features, especially inside the pupillary zone. Based on texture analysis, local spatial pattern of the iris consists of orientation and frequency information. Frequency information is discriminating in recognizing irises from different people [7].1D Log Gabor filter is used to extract the frequency information which represents the iris textures. A Log Gabor filter is a Gaussian transfer function on a logarithmic scale [5]. It has strictly band pass filter to remove the DC components caused by background brightness. The 1D Log Gabor filter on the linear frequency scale has a transfer function as in Equation (5).)))/(log(2/))/log(exp(()(2020w k w w w G −= (5)where w 0 denotes the filter’s centre frequency and k denotes the bandwidth of the filter. The ratio k/w 0 is set to be constant to generate consistent filter shape [6].Firstly, 2D normalized iris image is decomposed into a few 1D intensity signals. The 1D intensity signals are convolved with the 1D Log Gabor filter in Equation (5). 1D intensity signals are used because the information density is the highest in the angular direction, which corresponds to the horizontal row in the normalized iris images [7].The iris feature is extracted based on texture analysis. It is observed that the inner zone Z1 contains finest iris texture. The variations of the fine textureindicate that it contains high frequency components. The high frequency information can be extracted using Log Gabor filter with high center frequency, w 0. The middle zone Z2 has larger block of texture due to the presence of collarette boundary. It is processed using filter with a lower center frequency. The flattest texture appears in the outer zone Z3. The texture has low frequency components and therefore a coarsest filter with lowest center frequency is used to capture the local details of the outer zone Z3.After convolution, a series of real and imaginary numbers is generated. The phase information is quantized into four quadrants in the complex plane. Each quadrant is represented with two bits phase information. Therefore, each pixel in the normalized image is demodulated into two bits code in the template.The phase information is extracted because it provides the discriminating information for recognizing irises from different people. It does not depend on extraneous factors, such as illumination and imaging contrast.5. Template matchingHamming distance is defined as the measure of dissimilarity between two binary templates [2,3]. A value of zero would represent a perfect match. The two templates that are completely independent would give a Hamming distance near to 0.5. A threshold is set to decide the two templates are from the same iris or different irises.The fractional hamming distance is sum of the exclusive-OR between two templates over the totalnumber of bits. Masking template is used in the calculation to exclude the noise regions. Only those bits in the templates that correspond to ‘1’ bit in themasking template will be used in the calculation.maskBmaskA maskBmaskA templateB templateA HD I I I )(⊗=(6)Total hamming distance is summation of Hamming distance from three different zones with different weightings.321HD HD HD THD γβα++= (7)where HD i , i=1,2,3 denotes the Hamming distance between two templates computed from three different zones, Z1, Z2 and Z3. α, β and γ represent the weightings of the Hamming distance for zone Z1, Z2,and Z3 respectively. The weightings must satisfy the condition defined in Equation (8).1=++γβα (8)α, β and γ have decreasing weightings because inner zone provides more texture information than the outer zone. Zone Z1 contains most significant texture features that contribute to the recognition. Outer zone has less discriminating information and is often occluded by eyelids and eyelashes.In order to account for rotational variance, one of the two templates is shifted right and left bit-wise during matching. Each bit shifting in the template corresponds to rotation of the iris by an angle depends on the angular resolution. A few Hamming distances are calculated from successive shifts of the template. The lowest Hamming distance is chosen as the best match between the two templates. Finally, a threshold is set to distinguish whether the templates are from same iris or different irises.6. Experimental results and discussions6.1. Experimental resultsThe proposed algorithm was evaluated on CASIA iris image database version 1.0 [1]. There are 756 iris images from 108 different irises. For each eye, 7 images are captured in two sessions. The resolution of the iris image is 320×280 pixels. Figure 8. ROC curve for iris recognitionresults.ROC curve is plotted to measure the recognition accuracy. From the experimental results, the algorithm shows an overall accuracy of 98.62%. It is noted that the result is not perfect due to low quality of the iris images. The iris region is heavily occluded by eyelidsand eyelashes or distorted much due to pupil dilation and constriction. Some of the iris images are in defocused or motion blurred condition as shown in Figure 9. Image quality assessment is needed to select clear images with high quality.(a) (b) (c)Figure 9. (a) An occluded eye. (b) A defocusedeye. (c) A motion blurred eye.6.2. DiscussionsThe binary templates are encoded by quantizingthe phase information. Phase information isdiscriminating for recognizing irises from differentpeople. The templates contain no amplitudeinformation of the irises. Actual iris images cannot bereconstructed from the templates. Therefore, the use ofbinary templates could avoid the visibility of individualiris images.Furthermore, the compact templates are encoded inbinary format. The binary templates can be stored andprocessed effectively using DSP technology. Thousandof comparisons between different templates can becomputed within one second. Because the matching iscomputationally efficient, it is suitable for comparisonsof million of templates in large databases. Thisadvantage lets binary templates suitable forimplementing iris recognition devices based on DSPtechnology.7. ConclusionsAn innovative iris recognition algorithm based ontexture analysis is presented in this paper. The irisregion is divided into three zones according to thecharacteristic of the iris texture. Specific filters withdifferent center frequency are defined to extract thefrequency information from each zone. Differentweightings are given to each zone depending on itscontribution to the recognition. The encoded binarytemplates are small in size and able to avoid thevisibility of actual iris images. The binary templates aresuitable for implementing iris recognition devices usingDSP technology. The experimental results show thatthe proposed iris recognition algorithm is effective.The approach has achieved a high recognition rate upto 98.62%.8. AcknowledgementsThe author would like to acknowledge Institute ofAutomation, Chinese Academy of Science forproviding CASIA iris image database [1]. Thisresearch is partially funded by Malaysia MOSTIScienceFund 01-02-11-SF0019.9. References[1] “CASIA Iris Image Database,”/, 2007.[2] J. Daugman, “High Confidence Visual Recognition ofPersons by a Test of Statistical Independence”, IEEE Tans.Pattern Analysis and Machine Intelligence, vol.15, pp.1148-1161, 1993.[3] J. Daugman, “How Iris Recognition Works”, IEEE Trans.CSVT, vol. 14, no. 1, pp. 21 – 30, 2004.[4] R.P. Wildes, “Iris Recognition: An Emerging BiometricTechnology”, Proceedings of the IEEE, vol.85, pp.1348-1363, 1997.[5] D. Field, “Relations between the statistics of naturalimages and the response properties of cortical cells”, Journalof the Optical Society of America, 1987.[6] X. Yuan and P. Shi, “Efficient iris recognition systembased on iris anatomical structure”, IEICE ElectronicExpress, vol. 4, no. 17, pp.555-560, 2007.[7] L. Ma, T. Tan, Y. Wang, and D. Zhang, “Personalidentification based on iris texture analysis”, IEEE Trans. OnPattern Analysis and Machine Intelligence, vol 25, no.12, pp.1519-1533, 2003.。