A static video summarization approach via block-based self-motivated visual attention scoring mechanism

Since automatic visual semantic comprehension of video content is currently infeasible and unintelligent, key frames extracted from videos are inconsistent with human visual understanding. In this paper, a block-based self-motivated visual attention scoring mechanism named the BSVAS mechanism is proposed for extracting key frames. The approach described in this paper first reduces the dimensionality of the video by exploiting entropy as a static global characteristic measurement. Next, two block-based motion metrics are employed to express features from a spatiotemporal perspective, and a novel self-motivated strategy is applied to conduct feature fusion. Finally, a self-motivated scoring algorithm is performed to evaluate content attractiveness and frame importance to generate key frames. Experiments on gesture videos with various postures demonstrate that key frames extracted using the proposed method provide high-quality video summaries and cover the main content of the gesture videos as compared to several other excellent mechanisms in the literature.