Technology Research On Video-face Recognition Based On Class Specific Hyper Graphs

Face recognition is one of the most important research directions of computer vision,in recent decades,face recognition technology is developing rapidly and has obtained lots of inspiring results.While,because of the complexity of the problem itself,face recognition is still a challenging task.How to detect and recognize interested face target under the condition of less prior knowledge and complex imaging environment is the key problem for real-time video-face recognition system to solve.In this paper,we try to address a fundamental issue that how to recognize and tag the face in real-time videos with sample set of images or video collections and identification information.According to this basic problem,we put forward a video-face recognition method based on Adaboost-CSHG framework,and realized automatic face grouping based on attribute graph matching.At last,we set up a testing platform to verify the feasibility of the methods.The main works that we did in the paper laid out as follows:(1)Face ROI grouping.While modeling sample images,we should firstly group all the images,and then associate each group with the relevant flag information such as name or ID,thereby laying the foundation for the subsequent recognition and tagging.Besides,we often need to save video data in real-time video surveillance systems.However,we only tend to focus on the characters in these video files,because of the existence of background information in video images,storing the whole video file may cause large storage waste.If face ROI could be clustered before saving,much storage space will be stinted.In response to these requirements,we brought forward a face grouping method based on attribute graph matching and did some experiments to validate its effectiveness.(2)Video-face recognition based on CSHG.Due to dramatic imaging conditions change of video images,fixed face templates can't meet the demand of real-time video-face recognition.In this paper,we built attribute graph with SIFT features and its geometric constraint relationship,and all the image targets are characterized by attribute graph.We actualized face target tracking and on-line identification based on RSOM fast matching and CSHG self-learning.This face recognition way has strong adaptability to target imaging conditions change,which can satisfy the need of real-time video face recognition.In addition,many experiment results proved that the recognition performance of CSHG can achieve further improving through illuminationprocessing and RootSIFT.(3)Video-face recognition system design and implementation.According to the system functional requirements analysis and demonstration of the basic principle and solution,we advanced a face detection-and-recognition integrated system based on CSHG model.Moreover,the hardware and software design of the system is presented,and we also completed the design of each functional module,finally we carried out several simulations on the testing platform to verify the rationality and validity of the basic principles in this project.