Improved delaunay graph based video summarization with semantic features and canonical correlation

Key frame based video summarization, which enables an user to access any video in a friendly and meaningful way, has emerged as an important area of research for the multimedia community. Various pattern clustering techniques are applied for the extraction of key frames from a video to form a storyboard. In this work, we improve existing Delaunay graph based video summarization framework with i) semantic features represented by visual bag of words and ii) an improved feature fusion strategy with canonical correlation. Performance of the present method is compared with previous Delaunay graph based key frame extraction algorithms using Fidelity, Shot Reconstruction Degree and Compression Ratio. Experiments on standard video datasets clearly indicate the supremacy of the proposed approach.