Kavita Singh, Mukesh Zaveri, Mukesh Raghuwanshi
September 25, 2013
Abstract
This paper presents a new log Gabor-FPLBP-SVD (LGFSV) face representation technique that extracts singular values from log Gabor_FPLBP response image to form the LGFSV feature for face representation. The proposed LGFSV is invariant to changes in pose, illumination and facial expression. The novelty of this paper comes from (i) the design of minimal number of log Gabor filters to cover all directional shape features from face image which is further applied with Four phase Local Binary Pattern to enhance texture features in all direction; (ii) the extraction of singular value from each local matrix of log Gabor_FPLBP response image to form feature for face identification using nearest neighbour classifier; and (iii) extensive performance evaluation. In particular, the performance of the proposed LGFSV for face identification under pose variation and change in illumination and expression is evaluated on standard face databases such as ORL; Head Pose Image Database, Georgia Tech Face Database, CMU-PIE, GTAV and RLCI face databases. Experimental results with LGFSV show a significant improvement over individual face representation techniques.