The effect of Color space on discriminating power of Color local texture feature for Color face recognition

Abstract

Color local texture features (CLTF), proposed by Choi et al., exploit the discriminative information derived from spatiochromatic texture patterns of different spectral channels within a certain local face region to maximize the complementary effect taken by using both color and texture information for face recognition. Previous comparative experiments show that the CLTF extracted from ZRG and RQCr color spaces yield better recognition rates than FR approaches using only color or texture information. Nevertheless, it has been revealed that different color spaces have distinct characteristics, and thus effectiveness, in terms of discriminating power for the task of visual classification. Hence, in this research, we conduct extensive and comparative experiments to evaluate CLTF extracted from many different color spaces on four data sets, namely Color FERET, AR, SCFace, and Postech01. The results show that their performance is not consistent on different databases. This raises the need to develop a framework of choosing components from existing color spaces for the purpose of enhancing CLTF's discriminating power.