A Psychological Adaptive Model For Video Analysis

Extracting key-frames is the first step for efficient content-based indexing, browsing and retrieval of the video data in commercial movies. Most of the existing research deals with "how to extract representative frames?" However the unaddressed question is "how many key-frames are required to represent a video shot properly?" Generally, the user defines this number a priori or some heuristic methods are used. In this paper, we propose a psychological model, which computes this number adaptively and online, from variation of visual features in a video-shot. We incorporate it with an iterative key-frame selection method to automatically select the key-frames. We compare the results of this method with two other well-known approaches, based on a novel effectiveness measure that scores each approach based on its representational power. Movie-clips of varying complexity are used to underscore the success of the proposed model in real-time