Face Identification via Strategic Combination of Local Features

Abstract

Face identification systems that use a single local descriptor often suffer from lack of well-structured, complementary and relevant facial descriptors. To achieve notable performance, a face identification system is presented, which makes use of multiple representations of a face by means of different local descriptors derived from Local Binary Pattern (LBP) and Local Graph Structure (LGS). Then, max, average and L2 pooling operations are applied on each representation of face image to scale down the features. Further, different representations are combined together and formed a strategically concatenated feature vector. Three classifiers including two correlation based and k-nearest neighbor (kNN) are used to produce matching proximities which are used to characterize a probe user with rank identity. The produced ranks are then fused together using rank level fusion techniques. The experimental results determined on the Extended Yale face B database and the Plastic Surgery database are encouraging and convincing.