Improving verification accuracy by synthesis of locally enhanced biometric images and deformable model

In this paper, we propose a 2-stage preprocessing framework which consists of image enhancement and deformation techniques to increase the verification performance of image-based biometric systems. In the preprocessing framework, first the quality of biometric image is enhanced and then a deformation model is applied to minimize the variation between the two images to be matched. The proposed SVM image quality enhancement algorithm selects good quality regions from different globally enhanced images and combines them to generate a single high-quality feature-rich image. The proposed deformation algorithm is based on phase congruency information and locally minimizes the variations between two images while retaining the features required for recognition. The proposed algorithms are validated using face and iris biometrics as the two case studies. For performance evaluation, different face and iris recognition algorithms are chosen and the verification accuracy is computed using non-homogenous face and iris databases. Experimental results show that the performance of face and iris recognition algorithms are significantly improved when the input images are preprocessed using the proposed enhancement and deformation algorithms.