Local Bi-Directional Differential Exciation Co-Occurrence Pattern for Face Image Retrieval

This article presents a novel feature descriptor named as local bi-directional differential excitation co-occurrence pattern (LBiDECoP) for face image retrieval. The proposed method aims at extracting salient features present in face images using differential excitation and unique sampling strategy. The proposed feature descriptor fuses the human perception of pattern into feature extraction using differential excitation which mimics human visual system. Further, LBiDECoP encodes the co-occurrence of similar ternary edges for enriched discriminative power. The performance of LBiDECoP descriptor has been tested for image retrieval on two benchmark and challenging face image databases such as AT&T and FEI. The experimental results demonstrate that LBiDECoP outperforms the well known LTCoP, CSLBCoP, RICLBP and FDLBP descriptors in terms of average precision and average recall rates.