Abstract

Video summarization is a method to generate succinct version of a video by eliminating the redundant frames.  The representation of video summaries using key frames is a simple and effective way to generate video summaries. However, eliciting the frames that effectively characterize a video is a daunting task. A popular way to extract key frames is to compute the frame difference between the consecutive frames and then labeling a frame as key frame if a significant difference is located. In this paper, we propose a novel framework in which multiple index features, obtained from video frames, are combined to describe the frame difference between consecutive frames. It is observed that certain frame difference features have more influence in generating a representative frame difference measure. Moreover, some features are more relevant than others in different video genres. Therefore, for each video genre, the weights of different features are pre-determined at training phase by indirectly utilizing the Relevance Feedback Mechanism. Fuzzy Comprehensive Evaluation has been used to evaluate the efficiency of a particular frame difference measure based on the users’ feedback about summaries and thus generating the weights of each measure. The framework is evaluated based on three popular frame difference measures including color histogram, correlation and edge orientation histogram. The experimental results, based on an objective evaluation criteria, show that our technique gives better results as compared to some of the other techniques in the literature.

Key words: Image processing, key frame extraction, video summarization, fuzzy comprehensive evaluation, relevance feedback mechanism.