Multiscale non-linear video content decomposition: Analysis and evaluation

In this paper, we theoretically analyze the efficiency of a multiscale non-linear video decomposition scheme, which is used for progressive retrieval and navigation of video sequences (hierarchical summarization). In particular, the efficiency is expressed as the difficulty for a user to locate a relevant video segments, while moving through different levels of hierarchy. It is proved that multiscale video organization is more efficient than the sequential one, even in the worst case where the organization is not performed on a visual content perception basis. However, much greater improvement is achieved in case that video organization is accomplished with respect to the human perception of the visual content based on visual content classification algorithms. Experimental evaluation of multiscale organization is presented with respect to sequential scanning for different visual content organization and video decomposition methods.