Face Recognition Based on Multiple Video Cameras

Abstract

Face recognition is a fundamental task that enables advanced applications in video surveillance, human-computer interaction, and security. Existing methods of face recognition do not perform well on non-frontal faces, which often come from multiple surveillance cameras. This research aims to develop a method for partial face recognition, using images from multiple video cameras as a source and recognizing them against frontal images in a database. A key idea in the proposed method is to carefully evaluate a similarity between a set of video images from cameras and a frontal facial image from the database. We design two methods to evaluate the similarity. The first method directly measures the similarity from transformed video images. The second method fuses facial features of video images before measuring the similarity. We quantify performance of the proposed methods by comparing them with a competitive baseline using a public dataset. The comparison shows that the proposed methods have the highest recognition rates in four out of six test cases and have the recognition rates of thirty to seventy percent. The proposed methods of face recognition are promising and can recognize a face in difficult situations, including faces under occlusion or in a challenging environment.