17-03-2013, 04:09 PM
please mail me d seminar repots on this topic as soon as possible
17-03-2013, 04:09 PM
please mail me d seminar repots on this topic as soon as possible
25-06-2013, 12:31 PM
IRIS RECOGNITION PERFORMANCE ENHANCEMENT USING WEIGHTED MAJORITY VOTING IRIS RECOGNITION.pdf (Size: 213.03 KB / Downloads: 18) ABSTRACT Biometric authentication is a convenient and increasingly reliable way to prove one’s identity. Iris scanning in particular is among the most accurate biometric authentication technologies currently available. However, despite their extremely high accuracy under ideal imaging conditions, existing iris recognition methods degrade when the iris images are noisy or the enrollment and verification imaging conditions are substantially different. To address this issue and enable iris recognition on less-than-ideal images, we introduce a weighted majority voting technique applicable to any biometric authentication system using bitwise comparison of enrollment-time and verification-time biometric templates. In a series of experiments with the CASIA iris database, we find that the method outperforms existing majority voting and reliable bit selection techniques. Our method is a simple and efficient means to improve upon the accuracy of existing iris recognition systems. INTRODUCTION Biometric authentication has proven to be a reliable way to verify a human’s identity. The technology has certain advantages over more traditional password-, pin-, or hardware token-based human identification systems. First, biometric traits cannot easily be stolen, forged, or guessed. Second, there is no need to remember one’s biometric traits. Third, biometrics are difficult to repudiate. Due to these benefits, biometric authentication systems are being deployed in many real-world applications. Current systems employ many different biometric traits, including fingerprints, iris images, face images, retinal scans, palmprints, and gait patterns. IRIS RECOGNITION Iris Template Generation In an iris recognition system, the user presents his or her eye to an iris sensor, which images the user’s iris and generates a template from this image. Most iris scanners use near infrared illumination with normal monochrome CMOS or CCD camera sensors that are sensitive to near infrared light. After image acquisition, we use Masek and Kovesi’s algorithm [11, 12] for iris template generation, which is based primarily on Daugman’s methods [13]. Generating an iris template from a raw eye image involves three steps: iris segmentation, iris normalization, and iris feature encoding. Figure 1 shows the iris template generation process schematically. Verification Results To evaluate our weighted majority voting scheme, we performed two experiments, one without corrupted bit masking and one with corrupted bit masking. In both cases, we compared standard iris recognition (IR), reliable bit selection (IR-RB), majority voting (IR-MV), and weighted majority voting (IR-WMV) on CASIA. We computed Fisher’s ratio, decidability, and the ERR for each experimental condition. For IR-RB, IR-MV and IR-WMV, we used the first 5 CASIA images from each subject for training and used the remaining 2 images of each subject for testing. For IR, we used first image of each subject for training and used the remaining 6 images of each subject for testing. EXPERIMENTAL EVALUATION Iris Database and Algorithm We used CASIA version 1 [6] for our experiments. It consists of 7 iris images captured from each of 108 subjects, for a total of 756 images. The images were taken in two sets one month apart. We ran Masek and Kovesi’s segmentation algorithm [12] on each of the 756 CASIA images. The algorithm accurately located the pupil and iris in 83% of the images. We manually located the pupil and iris in the remaining images then performed the rest of the base template extraction procedure as described in Section 2.1. We should point out that CASIA version 1 has been criticized as an iris image database because the images have been altered to eliminate specular reflections in the pupil area [15]. However, although this makes the iris segmentation problem easier, in this paper, we treat iris segmentation and template generation as a black box. This means each of the bit weighting algorithms we compare benefit equally from the pupil alteration, and the relative performance of the bit weighting algorithms are not affected. Evaluation Criteria The accuracy of a biometric identification or verification system is usually measured in terms of its false acceptance rate (FAR) and false rejection rate (FRR), both of which depend on a distance or similarity threshold. To summarize these measures, we report the equivalent error rate (ERR), the point at which the FAR and FRR are equal. DISCUSSION AND CONCLUSION In this paper, we propose and evaluate a scheme for improving the accuracy of iris verification systems. As opposed to existing schemes, which either treat each bit equally or completely ignore unreliable bits, we use a distance measure that weights the bitwise comparisons according to the reliability of those bits at enrollment time. In two experiments on CASIA, one with corrupt bit masking and one without, we find that the method performs better than existing schemes using multiple enrollment-time scans to obtain more reliable templates. The method treats the template generation algorithm as a black box, so it can be used to improve the performance of any biometric verification system that employs bitwise comparison of binary templates. |
|