Clothing Change Aware Person Identification

Abstract

We develop a person identification approach - Clothing Change Aware Network (CCAN) for the task of clothing assisted person identification. CCAN concerns approaches that go beyond face recognition and particularly tackles the role of clothing to identification. Person identification is a rather challenging task when clothing appears changed under complex background information. With a pair of two person images as input, CCAN simultaneously performs a verification task to detect change in clothing and an identification task to predict person identity. When clothing from the pair of input images is detected to be different, CCAN automatically understates clothing information while emphasizing face, and vice versa. In practice, CCAN outperforms the way of equally stacking face and full body context features, and shows leading results on the People in Photo Album (PIPA) dataset.