A NOVEL TEXTURE-BASED MULTI-LINEAR ANALYSIS ALGORITHM FOR FACE RECOGNITION

Abstract

In this paper, a multi-linear approach based on texture features for face recognition is proposed. First, we extract fragment-based texture features of the facial images using the local binary pattern (LBP) descriptors, which capture both shape and texture information and also are robust to illumination variations. Second, we propose high-order orthogonal iteration (HOOI) algorithm that obtains optimum truncated factor-specific modes, which are not guaranteed in the standard N-mode SVD algorithm, in an iterative manner. Finally, we apply HOOI to obtain a compact and effective representation of the facial images based on the texture features. Our representation yields improved facial recognition rates relative to standard eigenface, tensorface, and other popular algorithms, especially when the facial images are confronted by a variety of viewpoints and illuminations. To evaluate the validity of our approach, a series of experiments are performed on the CMU-PIE facial databases.