Evaluation of LBP-based face recognition techniques

Face recognition is a popular technique in identifying human features. In certain application such as recognizing criminals from video surveillance, where no other physical trait is available, face recognition is the most practical and assessable human recognition method. For this reason, face recognition continue to attract large research interest among image processing community. In this paper, Local Binary Pattern (LBP) texture method is used to characterize the image features. Four derivatives of LBP are evaluated in order to select the best LBP technique for face recognition system. The derivatives are conventional LBP, Center Symmetric Local Binary Pattern (CS-LBP), Local Binary Pattern Variance (LBPV) and Completed Local Binary Pattern (CLBP). The evaluations of the LBPs are conducted using Japanese female facial expression (JAFFE) and author personal databases using recognition rate and run time value as the performance metrics. In particular, three different experiments are conducted, namely LBPs in an ideal environment, LBPs in different level of contrast and LBPs in the presence of additive Gaussian noise. The results indicates that based on average recognition rate, the LBPV gives the best performance among the LBPs and consider as the most reliable LBP derivative in change of illumination and noisy environments.