Video Retrieval System Based On MPEG-7 Low-level Features

With the rapid development of the multimedia technology and computer network, digital video showing explosive growth, it's very concerned in info-field that how to organize ,manage and retrieve multimedia information efficiently and find the exactly video in need quickly. The traditional video retrieval method is text-based, you must annotate the video with text firstly. This method takes too much time and in many case you can't describe the video information properly using text. Therefore, in order to efficiently analyze and retrieve the video data, CBVR(content-based video retrieval) technology has been proposed, it's main idea is retrieving the video data based on low-level features and semantic features. Compared to the traditional text information retrieval, video data information is too large and hard to describe, so theMPEG-7(Multimedia Content Description Interface) standard is proposed. MPEG-7 is an ISO/IEC standard developed by MPEG (Moving Picture Experts Group), is describing the multimedia content data that supports the metadata and text description of video content and provides a description and retrieval specifications. With the progressive development and refinement of MPEG-7 standard, the CBVR technology and system had been much more promoted.This paper combined MPEG-7 color, texture and shape feature to analyze the video key-frame. First, extract the key-frame from the video clip and extract the MPEG-7 descriptors feature vector from fey-frame, such as color, texture, shape et al, then stored feature vector in the database as index for content-based video retrieval.Because the key-frame feature vectors are all high-dimensional vector, if we use the sequential scan method, the retrieval speed will be unacceptable. Therefore, this paper study the high-dimensional indexing method, similarity measurement of video clip and video clip retrieval algorithm, and decided to use the BitMatrix high-dimensional algorithm to index the feature vectors, which greatly improve the retrieval efficiency of the video retrieval system.Experiment results show that efficiency of the video clip retrieval system designed in this paper is much faster than the sequential scan algorithm, while its retrieval precision can meet the user's requirement.