Video Super-resolution Reconstructionwith Application To Face Recognition

In the past few years, the problem of public security problem becomes severe due to the economic development. In order to react timely in emergency, and to provide clues for cracking criminal cases, cameras have been widely used in public places such as plaza, sidewalks and communities. As a consequence, to identify people from the mass quantities of video obtained by public cameras is an important but exhausting task. Automatically face recognition by computer, because of its speed, efficiency, effectiveness and the potential capability to solve the problem mentioned above, has become a hot topic. Face recognition technique is an important ramification from recognition techniques based on biological characteristic, and a frequently discussed problem in the realm of pattern recognition and computer visualization.A complete face recognition system is comprised of image enhancement, face detection and face recognition. For those systems that are customize to public areas, the resolution and definition are usually too low to recognize people's face effectively. Current face recognition systems commonly lack the enhancement solution to process images with low-definition. This paper takes a practical approach to discuss the primary algorithms and its features for each component in face recognition technique, and apply super-resolution reconstruction technique innovatively to face recognition systems. This research designed an improved super-resolution algorithm based on multi-video-frame wavelet fusion, and it provides the first design that combine this technique and canonical face recognition algorithm to construct a complete face recognition system.First, this paper summarized the prevailing algorithms related to each component of face recognition system, and then discussed the theoretical foundation of image super-resolution and a variety of interpolation algorithm. Upon discussing the single frame super-resolution algorithm and POCS algorithm, this paper propose a new applied super-resolution algorithm and make a experimental comparative study. In order to construct a face recognition system, we discuss these issues in image pre-processing: filtering, histogram-balancing, motion detection and space transform, and then we focus on the Chen-Vese algorithm in the part of face detection. As to the part of face recognition, we discuss three canonical algorithms: BP network, Gobar and PCA. Base on the discussing of these algorithms, this page describe the design and implementation of a new face recognition system integrated with the capability of image super-resolution.This research is oriented in user requirement. Based on our analysis of the features of monitoring video, the system can cut up pictures and reconstruct it using motion detection, which is proved to be an effective approach for face detection and can achieve a better outcome of super-resolution technique. Under the guidance of low-coupling-design, we conducted an experimental comparative study of different face recognition algorithm such as Gobar and K-L transform, and get results of both with and without the image super-resolution. The results turned out to support our hypothesis that the new face recognition system can achieve better detection rate and recognition rate.Above all, face recognition systems is a sophisticated problem which involving issues from multiple field. This paper gain some new insight in the realm of face detection and recognition, and analyzes the gap between face recognition systems and its practical usage, and finally propose the target that need to be work towards in future research.