A Research of Face Super-resolution Technology in Low-quality Environment

In recent years, with the extensive application of video surveillance system in police’s criminal investigation, video investigation technology is playing a more and more important role in detection of cases. Key characters in case spots, such as criminal suspects and witnesses are often targets that attract most attentions. Therefore, face image become the key clue of a case. While in actual surveillance, due to distance and limitation of equipment, definition of most face images is of low-quality. Face superresolution technology can make use of information in face sample database to restore high definition face image, which can effectively enhance definition of face images in surveillance video and restore detail information of face features. This technology is extremely important to improve the clarity of face image, increase accuracy of face recognition, and thus increasing the police’s ratio of solving cases. In this thesis, the author introduces shape semantics model and posterior image information into the frontier global face and partial face super-resolution approach, and put forward the corresponding algorithm, which can not only increase the robustness of single mode face super-resolution algorithm, but also can be used for reference for other image processing technology that research robustness. Simultaneously, this thesis extend the approach based on shape semantics model and posterior image information to multi mode face super-resolution, which increases the robustness of multi mode face super-resolution, and provides train of thought for applications of the approach put forward.