Heterogeneous Face Identification by Fusion of Local Descriptors

Abstract

This paper reports a novel approach for heterogeneous face identification which makes use of fusion of MB-LBP (Multi-Block Local Binary Pattern) and MLBP (Modified Local Binary Pattern) descriptors at feature extraction level. Face images obtained from different sources such as visible, sketch (viewed and forensics) and NIR (Near Infrared) face images are used for the experiment. In order to achieve robust heterogeneous system, MB-LBP and MLBP are applied to the individual heterogeneous face images for extracting local features from visible, sketch and NIR face images. These local features are used to capture the structural description of a face image. Feature vectors obtained from these two local descriptors are fused together by concatenation and fused feature vector is passed through the classifier. The experimental results are determined from two well-known heterogeneous face databases, namely, LDHF and IIIT databases. Rank order statistics is used to determine the ranks of the face images according to their identification probability. Experimental results for identification in heterogeneous environment are found to be very consistent and convincing.