Design of Video Retrieval System Using MPEG-7 Descriptors

Abstract

The paper proposes a content-based video retrieval system designed using MPEG-7 (multimedia content description interface), which provides a standard description for a video. The system consists of three parts: shot boundary detection, feature extraction and similarity measurement. In shot boundary detection, cut and dissolve can be detected using the histogram difference and skipping image difference, respectively. In feature extraction part, two MPEG-7 visual descriptors, Color Structure Descriptor (CSD) and Edge Histogram Descriptor (EHD), are used to represent the color feature and edge feature of the key frames. Lastly, the similarity between key frames is calculated using dynamic-weighted feature similarity calculation. The proposed system is tested on three kinds of videos. Promising results are obtained in terms of both effectiveness and efficiency.