Research On The Quality Evaluation Of The Sequence Iris Image

The primary issue of iris recognition is to obtain the iris image. there are twoways in general: (1) capturing one iris image once statically, and then carrying onprocessing to this image (2) Capturing certain images in the certain time by meansof video frequency flow and choosing one or several images which can meet thequality requirement to use in following processing. The capturing equipmentdesigned independently by our laboratory belongs to the second way. In order tocapture the iris image automatically and completely in this way, a good qualityevaluation method of the iris sequence image is needed.This paper carried on the discussion and the research on the pretreatment andevaluation method of the clarity and the quality of the single frame image thenestablished the sequence image quality evaluation process by synthesizing theresult of the single frame image and the characteristic of the sequence image.The pretreatment of single frame image includes the iris localization, theeyelid detection, the eyelash detection, rotation processing, the iris normalizationand the image enhancement.According to the nature of HVS(Human Vision System), ROI (region ofinterest) is one kind of positive feeling behavior of the vision, which is not onlyrelated to physiological factor, but also is decided by the psychological factor togreat extent. The entire image quality is basically decided by the quality of "ROI",but the quality of region of not interest had little impact on the whole quality. Theresearch in this paper fully used the ROI theory, according to the different stagesand the different processing goals, many kinds of ROI tallies with the nature ofhuman vision system had been determined, which played a crucial role in solvingthe problem.The iris localization used two steps localizations method: inner circle islocated firstly and the outer circle is located later. Facula is located according to theimage facula characteristic from capture equipment in the process of capture and then asmall area whose center is the facula center is selected as location ROI.At last, theedge of such ROI is detected by Canny operator, then inner edge of the iris islocated by Hough transformation, and the center of the iris is obtained;on the baseof the view that the outer and inner circle of iris are concentric circles, after theoriginal image is enhanced, the outer circle of the iris is located by means of unilinegradient maximum value which passes through the center of a circle.The eyelid and the eyelash are the important interference source of iris;therefore the detection of eyelid and eyelash also has important impact on thefollowing recognition. In this paper, the interference conditions the eyelid and theeyelash to iris is identified by the ratio of the eyelid and eyelash area to the area ofiris region.In the captured eye images, the gray scale of eyelid is bigger than the gray scaleof iris region, that is, there is a comparatively big gradation gradient between theinterference eyelid in the iris region and the iris. So long as we extract the suitablequantity of the critical point corresponding coordinates, the eyelid boundary linescan be determined through the curve fitting, and the interference eyelid in the irisregion is noted. Then the eyelid area can be computed through area approachesmethod.The eyelash gray scale basically approaches the pupil gray scale, but is of greatdifference from other parts of the image gray scale. This paper made binarization tothe original image through the auto-adapted threshold value method, and the areaof interference eyelash is obtained through computing pixel point appears in the irisregion.The wavelet transformation is an analysis method combined time domain andfrequency domain. the image is decomposed on the time domain and the frequencydomain at the same time, and the multi-resolution data which produces is easy toconsider the humanity visual characteristic. The wavelet coefficient spatialdistribution has the very good corresponding relations with the original imagespatial distribution: The LL frequency band is the thumbnail of the image content,which is the frequency band where image data energy is capture;The HL frequencyband has reflected the variable information and the edge information in thehorizontal direction of the image;The LH frequency band has reflected variablegradation information and the image edge information in vertical direction of theimage;The HH frequency band has reflected synthetic variable gradationinformation of the image gradation in the horizontal and vertical direction. Thehigher the energy of frequency band is, the clearer this image is.After wavelet transformation, the LL frequency band is the samplingthumbnail nature of the image content In order to enhance the speed of the irislocation, this paper carries on the iris localization to the LL frequency band imageby using this nature, and then maps the parameter on the original image to obtainthe parameter of the iris center of a circle and radius of inner and outer circle.Theexperiment result proved that these parameters accorded with the parameterobtained in the original image when making the iris localization. The localizationtime reduces greatly.when Carrying on the clarity evaluation to the iris image, in order to avoidthe interference from the eyelid and the eyelash, this paper selects an area of 64×64on the left and right and 32×32 at the top and bottom of the iris region as theinterested area of clarity evaluation. These high frequency coefficients are analyzedafter wavelet transformations in the region, which is taken as the clarity factor ofiris image.The combination of interference factor and image clarity factor can be usedas comprehensive quality evaluation for a single picture. For an input sequenceimage, in order to reduce the image processing time during the total evaluationprocess, we come to the conclusion that the step-type screening from rough torefined is adopted after the research.Different thresholds are established in different stages, only the images thatmeet the need of the thresholds are able to enter the next evaluation stage. In thisway, evaluation time for the sequence image is reduced effectively, and the qualityand speed of selected images have been guaranteed. According to the irisrecognition is the single template or multi-templates way, the single frame or themulti-frames images conforms to the quality requirement are selected.The main work this paper completed is:(1) Established the evaluation model of the sequence iris image based on ROIof the human vision.(2) The inner and outer circle of the iris is located quickly based on thelocation ROI which is the characteristic facula produced by the capturingequipment and the wavelet low frequency area which has the sampling thumbnailnature of the original image. According to the detection on the correlationcoefficient of pupil, we carry on the iris living specimen judgment, detect theeyelid and the eyelash area and obtain the interference factor.(3) Taking a located area of 64×64 on the left and right and 32×32 at the topand bottom of the iris region as the clarity to judge the ROI. With the combinationof wavelet transformation high frequency energy statistical analysis method, theimage clarity factor is obtained.(4) Realized the step-type screening of iris sequence image and the fast qualityevaluation software. The solid foundation for the embedding processing of futureautomatic iris capturing is laid down.