Font Size: a A A

Research On Key Technology Of Live Iris Recognition

Posted on:2008-04-30Degree:DoctorType:Dissertation
Country:ChinaCandidate:X F HeFull Text:PDF
GTID:1118360215476850Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
Biometric recognition or, simply, biometrics refers to the automatic recognition of individuals based on physiological or behavioral characteristics. Physiological characteristics, such as face, iris, fingerprint, palmprint, voice etc. are connatural, whereas behavioral characteristics which are the habits of human, such as gait, handwriting, are postnatal. Compared with traditional personal recognition methods, biometrics has the following merits: (1) memoryless; (2) difficult to forge; (3) useable whenever. Iris recognition is one of the most reliable biometrics in terms of identification and verification performance.In this dissertation, some of the key issues related to a live iris recognition system are investigated, including iris acquisition, iris image quality assessment, iris preprocessing, fake iris detection, iris feature extraction and matching. The main contributions of this work are as follows:First, one of the major challenges for automated iris recognition is to capture a high-quality image of the iris. Most of current capture devices need user's cooperation. To some extent, the interactive mode demands cooperation of the user who needs to be trained in advance and thus eventually increase the time of image acquisition. This dissertation introduces three kinds of iris capture devices. The previous two versions of automatic iris capture devices are contact capture devices, not convenient for users. The third version is a contactless auto-feedback iris capture system. We designed a LCD screen above the camera to realize the exact focus of iris image via the real-time feedback of the captured iris image in the LCD screen which makes it easy to use and non-intrusive for users. Iris image quality is not affected even if the user wears the glasses or sunglasses. Furthermore, three versions of iris database are also presented which are respectively created by using the three kinds of iris capture devices.Second, not all the iris images obtained from the capture device are of high quality and suitable for recognition. Some images, such as defoused image, motion blurred image, deformed pupil, eyelids or eyelashes occlusion seriously will influence iris recognition performance. Therefore, it is necessary to assess the iris image quality as much as possible. This dissertation analyzes and compares several representative quality assessment methods, and then proposes an effective method based on Laplacian of Gaussian operator for assessing the quality of iris image including defoused and motion blurred images. Since this method needn't localization, it is superior to the traditional quality assessment methods both in speed and accuracy.Third, iris preprocessing mainly includes iris localization, normalization, enhancement and noise detection etc. The most important part of the iris prepreocessing is the iris localization which includes inner circle (pupil), outer circle and eyelids localization. Iris localization is a time-consuming process. Moreover, the result of iris localization will influence the subsequent processing since iris pattern represented improperly will inevitably result in poor recognition performance. In this dissertation, a robust localization method is proposed. The pupil is detected using geometrical method. The shrunk image is used to detect outer circle based on modified Canny edge detector together with Hough transform. Moreover, ground truth is introduced for performing a quantitative analysis of the proposed algorithm's results. The upper and lower eyelids are modeled by two parabolas using the Least Squared Criterion and the direction of the parabola is not considered in this dissertation in order to speed up the algorithm. Experimental results show that the proposed method can estimate the eyelids contour with good accuracy.Fourth, fake iris detection is discussed in this dissertation. Fake iris detection is to detect and defeat a fake (forgery, counterfeit) iris image. The characteristics of the current different types of fake irises are analyzed detailedly, including iris photograph, printed iris on a paper, color contact lens, simulation eye etc. To overcome the shortcoming of the previous research methods, this dissertation proposed a novel method for detecting color contact lens based on statistical texture analysis and support vector machines (SVM) classification. We have found empirically that region of interest (ROI) is usually concentric with the outer circle of iris, and the radius of the ROI area is restricted in a certain range. So, the distinctive features of the ROI area are used for fake detection. SVM is selected to characterize the distribution boundary, for it has good classification performance in high dimensional space. The experimental results indicate that the new approach is better than traditional method both in speed and accuracy.Last, iris feature extraction and matching is discussed in this dissertation, which is the last part of an iris recognition system. It is difficut to extract iris feature and classify the feature exactly. This dissertation presents a new feature extraction method based on dual-tree complex wavelet transform (DT CWT). DT CWT does not only keep wavelet transform's properties of multiresolution decomposition analysis and perfect reconstruction, but also adds its new merits: approximate shift invariance, good directional selectivity for 2-D image, and limited redundancy, which are useful for iris feature extraction. Experimental results show that the proposed method is better than traditional method. Moreover, comparision results of different distance matching methods prove that the phase information of the DT CWT coefficients is more distinctive than the amplitude.
Keywords/Search Tags:Biometrics, Live Iris, Acquisition, Image Quality Assessment, Preprocessing, Localization, Fake Detection, Feature Extraction, Matching
PDF Full Text Request
Related items