Cross-Domain Face Recognition Using Dictionary Learning

Abstract

Cross-domain face recognition refers to the matching of face images between different domains. It has many applications in night time surveillance, border security surveillance and law-enforcement. However, this is a difficult task because of the non-linear intensity pixel values between the images which occur due to the domain gap. Recently, dictionary learning methods such as coupled dictionary learning and domain adaptive dictionary learning methods are used to solve this problem. In this paper, we propose a dictionary learning based method to learn the common subspace in order to reduce the gap between domains. Initially, we separate the domain specific representation and identity related representation by using commonality and particularity dictionary learning. In the next step, we remove the domain specific representation and get the common subspace. Thereafter, in order to get the more discriminate representation, we use metric learning. The proposed method is tested on RGB-D-T data set and the experimental results show that the proposed method is performing better even when there is no person common between training and testing sets.