Shape Based Automatic Annotation and Fuzzy Indexing of Video Sequences

Abstract

Multimedia content available on the web is increasing at a rapid rate. At this rate, indexing video content manually for search and retrieval is getting increasingly difficult due to the high cost and time that it demands. In this paper, we propose a scheme for automatic annotation and indexing of video content with a goal of supporting efficient retrieval. The framework performs the annotation and indexing by operating on the key representative frames of a video sequence. The proposed framework broadly involves three major steps. The first step involves pre-processing, segmenting a video frame into objects and obtaining the boundary coordinates. In the second step, shape descriptors are computed by applying discrete Fourier transform on the centroid distance function. In the third and the last step, template based matching and fuzzy indexing of the video is performed. Simulations were carried out for evaluating the performance of both pre-processing and annotation using the database-MPEG7 CE Shape-1 Part B. The results show that the framework could effectively perform annotation.