Design And Implementation Of Target Face Query System For Video

With the continuous development of intelligent information technology and the widely application of multimedia technology, more and more information is collected and recorded by video. At the same time, however, the enrichment of video information brings some problems such as how to quickly retrieve the needed contents within large amount of video, including screening the criminals or looking for the lost old men or children in the surveillance video, searching for the target person from the network video, etc. So in order to achieve the purpose of saving manpower and improving efficiency, a research should be done to achieve the function of video retrieval based on content.After researching some relevant technical on video-based face retrieval from scholars at home and abroad which is combined with the relevant requirements of this system simultaneously, this thesis uses C# language which is highly efficient and flexible to design and develop a target face query system for video in the environment of Visual Studio 2010 and EmguCV. This system regards human face as the searching object, and achieves many functions including multi-face detection based on image and video, face extraction, face matching, matching fragments integration and fixed-point video playback.Firstly, This system extracts key frames to do the preprocessing operations, and then uses the AdaBoost algorithm which is fast and suitable for real-time face detection to detect and extract face from reading images and video frames, then matches the detected face with the target face by SURF algorithm and records the time point of the current video frame if they matched. SURF algorithm has scale-invariant feature and its matching speed is faster than others, so it can be applied to the face recognition, image registration and many other fields. Finally, this system integrates and outputs the matched video sections, and realizes freely playback between sections through DirectX. It is convenient for staff to further observation, and improving browsing efficiency.This thesis chooses the NRC-ⅡT video face database, life video collection and TV video clips on to test the target face query system. The experimental results show that the system can accurately retrieve the appearing time of the target face, and it is also easy to operate and has a certain stability and practicability. This system can be used to relieve the heavy work of checking the videos to search target face artificially, and can efficiently reduce the workload of artificial identifying target face.