Research And Implementation Of Key Technology For Fast Video Retrival

Information technology has been developed rapidly and multimedia video has been widely used.Video can save a lot of information.The explosive growth of video has caused a series of problems in the rapid developing society.Video-based information retrieval has many applications.Video-based face retrieval is one of the valuable research directions.In recent years,how to find target faces accurately and efficiently from mass videos has become a hot topic in the field of video retrieval.At the same time,the research also promotes the development of video-based face recognition,image processing and large-scale image retrieval.In this thesis,we have studied the video based face retrieval technology,focusing on video key frame extraction,face detection,image processing and face recognition.An improved iterative quantization hash algorithm is proposed for the matching of similar face retrieval.Finally,a video based face retrieval prototype system is developed.The main work of this article is four parts:(1)Key frame extraction: I have studied the mainstream key frame extraction methods,such as camera-based,content analysis,motion analysis,clustering,compressed video and sampling.In a comprehensive comparison,the sampling method is selected for video key frame extraction.(2)Face detection and location: We studied the mainstream video face detection algorithm,focused on the AdaBoost face detection algorithm based on Haar feature and the Libfacdetection algorithm based on MB-LBP.Through experimental comparison,the Libfacdetection algorithm was finally used to detect and locate the static face images and video frames.And I have extract the coordinates of the 68 feature points of the human face as the identified features.(3)Image processing: face detection in video is often affected by the difference of illumination,image background,image specification and face attitude.In this dissertation,the detected facial images are preprocessed including image enhancement,noise removal,face alignment,scale normalization and so on.(4)Face recognition and retrieval: This thesis presents an improved iterative quantization locality sensitive hashing algorithm for approximate face retrieval,and we have done some experiment between the improved iterative quantization hashing algorithm and other mainstream local sensitive hash algorithm to compare retrieval accuracy and retrieval rate.The comparison results show that the improved iterative quantization hash algorithm improves the retrieval accuracy and speed effectively.Finally,I applied the improved iterative quantized hashing algorithm to a similar face retrieval module of video face detection system.