A NEW BILINEAR APPROACH FOR INCREMENTAL VISUAL LEARNING AND RECOGNITION
Haïfa Nakouri, Mohamed A. Limam · International Journal of Pattern Recognition and Artificial Intelligence · 2013
The intra-subject variation of a captured image under different style conditions can be much larger than the inter-subjects variation. This fact highly affects the recognition rate. Among approaches in the literature, bilinear models are efficient two-factor models that separate observations into two factors namely style and content. This paper proposes a new incremental robust face recognition method based on separating the identity factor and the style factor using a symmetric bilinear approach. The proposed method uses an iterative orthogonal triangular factorization for bilinear model learning, along with a translation procedure of the latter. This translation requires a repetitive computation of the orthogonal triangular factorization to reach the content and style factors of an unknown subject. We perform experiments with six databases in terms of recognition rate and style accuracy under four different styles: variety of illuminations, head positions, face expressions and occlusions. Experimental results show that our approach achieves better performance in terms of recognition rate than other existing methods such as the bilinear model, the ridge regression bilinear model and the Robust Principal Component Analysis.