Abstract

In this paper, a sub-image method for face recognition combined with Local Binary Patterns (LBP) and recursive methodology is proposed, which looks forward to improve the robustness against local variations and reduce the runtime for Random Subspace Method. Firstly, LBP is employed as a preprocessing step prior for the face images. Then, the preprocessed images are divided as a sub-image set by the sub-image method. Finally, a Random Subspace Method is performed in sub-images by a new random sampling (called as recursive random sampling). Experimental results on the ORL and Yale face databases reveal that the texture features extracted by LBP are more discriminative than the original pixel features. And the recursive random sampling method can improve the operational efficiency while achieving a comparable recognition accuracy.

Paper Details