Interval type-2 fuzzy based pixel wise information extraction: An improved approach to face recognition

Over the last few decades, face recognition has become a popular area of research in computer vision. It is the most successful application of pattern classification in the field of image analysis and understanding. Despite of all success in face recognition systems, it faces the problem of handling inaccurate information. In this paper, we propose a fuzzy logic based information extraction approach for face recognition systems for diminishing the effects of uncertainty formed by various reasons like variations in light direction, facial expressions, face poses etc. To address the issue of uncertainty, interval type-2 membership function on each random generation of train and test image vector is applied. Further, we perform dimension reduction of high dimensional image space using principal component analysis. Next, we apply one of the variant of distance metrics i.e. k-nearest neighbor classification, to obtain the classification error rate between and within class scatter matrix. Type-2 fuzzy set membership function employment makes the system able to reduce average error rate more and, also it handles uncertainty more effectively as compare to type-1 fuzzy set membership function. Experiments performed on the AT&T database validate our proposed approach and the proposed approach does not require any pre-conditioning.