Robust Face Recognition with Partial Distortion and Occlusion from Small Number of Samples Per Class

The posterior union decision-based neural network (PUD-BNN) has been proposed in our previous work for dealing with face recognition task subject to partial occlusion and distortion. However, one difficult of this method is inaccurate to model classes with only a single, or a small number of training samples. In this paper, we proposed an extern approach to tackle above problem by two strategies. Firstly, the new approach artificially constructs some new training data with original training images for complementing training data. Moreover, an efficient density estimation method is used into PUDBNN to tackle the reliable likelihood densities estimation with insufficient training samples. The new approach has been evaluated on two face image databases, XM2VTS and AR, using testing images subjected to various types of partial distortion and occlusion. The new system has demonstrated improved performance over other systems acronyms.