Face Verification Between ID Document Photos and Partial Occluded Spot Photos

Abstract

ID-spot face verification is an important problem in face verification area, which aims to identify whether the spotted face is the same to the ID photo. Although some face verification systems have been deployed in many application scenarios, most of them are used in a constrained environment and many key problems need to be addressed furthermore. In this paper, we focus on a challenging ID-spot face verification task, in which the spot photo is partially occluded. Toward this end, a two-stream network is employed to learn more discriminative feature for distinguishing different ID-Spot face pairs. In addition, to suppress the negative effect of background and occlusion, a global weight pooling method is proposed, which makes the available face area more significant than the background and occlusion. The experimental results show that the proposed method obtains 10% improvements on FAR@0.01 compared with previous schemes.