Philosophy design of single-trait based multi-feature biometric system

Abstract

This paper presents new techniques for designing a simple and reliable multifeatured biometric system based on a single trait source. First, a one-to-one relationship between the feature’s edge and its associated angle is utilized after extracting the contrast feature using the gray-level co-occurring matrix (GLCM) method. Secondly, the classifying stage is modified to process one-dimensional vectors rather than the whole feature’s template. That means whatever the template size is, the matching operation is always processing a one-dimensional vector called a mean-feature vector which requires low storage and less computation complexity. Finally, for comparison purposes, the performances of the three biometric systems are calculated for recognizing 170 subjects taken from four facial databases. These comparisons are made using three error distance measurements. The recognition rates of the angle-based feature were very competitive to the regular edge-based results; however, the overall recognition accuracy is highly improved after fusing the decision of the two unibiometric systems using the Logic-OR operator. The fused system performance was satisfactory and it shows that the decision fusion of the single source trait based multifeatured system has promising performance represented by accuracy improvement, low storage, and low matching time.