Face Image Validity Assessment In Surveillance System

With the pervasiveness of video surveillance and the development of computer vision technology, people are trying to make more use of video surveillance systems. Recently, demands on the validity of face image in video surveillance system, like ?facial features need to be clear in surveillance videos?, are adopted by many national or local surveillance industry standards.This paper proposes a technology for face image validity assessment in surveillance system. This technology can quantify the validity of face image in surveillance video, and is able to replace human subjective assessment with a more fast and objective assessment implemented by computer vision technology. Therefore, the technology proposed can deeply bring benefit for scientific set and assessment of surveillance system, thus contributes to the development of surveillance.On one hand, the paper analyses face image validity trough plane feature. A subjective assessment data base for face validity is built firstly, which will be the benchmark for algorithm performance assessment. Then the influence that blur and white noise have on face image validity is analyzed and proved by experiment. Then, BRISQUE, a state of the art IQA algorithm is tried to assess noise of a face image, and experiments results proves that BRISQUE can be adopted to face image validity assessment to some extent. However, BRISQUE is not effective in assessing image entropy, which is important for face image. So this paper proposes a face plane feature assessment algorithm based on spatial features and image entropy, whose performance is then proved by experiment.On the other hand, to compensate for the miss of plane feature analysis, posture feature analysis is proposed by this paper. A face orientation recognition algorithm based on BP neutral network is proposed in this paper, which also includes a newly designed edge feature extraction methods. And this face orientation recognition algorithm is proved to have a better performance than previous research in a follow up experiment. Besides, a face image validity assessment system for surveillance is proposed, which not only includes the plane and posture feature assessment algorithms proposed in previous sections, but also includes a face detection system. And the experiment of the system execution also proves that the research of this paper has a great value for setting and assessing video surveillance system.