Deep representation for partially occluded face verification

Abstract

By using deep learning-based strategy, the performance of face recognition tasks has been significantly enhanced. However, the verification and discrimination of the faces with occlusions still remain a challenge to most of the state-of-the-art approaches. Bearing this in mind, we propose a novel convolutional neural network which was designed specifically for the verification between the occluded and non-occluded faces for the same identity. It could learn both the shared and unique features based on a multiple network convolutional neural network architecture. The newly presented joint loss function and the corresponding alternating minimization approach were integrated to implement the training and testing of the presented convolutional neural network. Experimental results on the publicly available datasets (LFW 99.73%, YTF 97.30%, CACD 99.12%) show that the proposed deep representation approach outperforms the state-of-the-art face verification techniques.