Integrated System for Face Detection, Clustering and Recognition

Recent years, many approaches have achieved remarkable performances in face detection, clustering and recognition. However, real life can barely see systems designed for their combination, and thousands of videos and images need to be handled in a timely manner. In this paper, an end-to-end face detection, clustering and recognition system has been proposed. The input of our system can be videos or images. For videos, we firstly utilize ffmpeg to transfer these collections into separated frames and for images this step can be skipped. Secondly, we employ joint detection and alignment to detect faces in frames. Then the clustering step is conducted to find face relations based on Interpretive Structure Model. Finally, we establish our own dataset and train our own classifier to realize face recognition based on FaceNet. Most important of all, we propose our original clustering method which avoids duplicate feature computation and repetitive face recognition. And it enhances the efficiency of handling data and shortens the runtime of our experiments greatly. This system is evaluated using our designed database of 43 persons' faces with varying scales and poses obtained on different complex backgrounds. The performance of the system is quite good and it achieves average accuracy of 87% to 92%.