Video Abstraction via Attention Model and On-Line Clustering

Video abstraction is an indispensable component in various applications, such as indexing, browsing and retrieval. In this paper, we present a new video abstraction algorithm based on visual attention model and on-line clustering. Representative frames are first selected on shot level. The attention regions in representative frames are detected via attention model. Finally, the visual features of attention regions are clustered in an on-line manner to reduce memory cost. Experimental results demonstrate that the key frames extracted by the proposed algorithm are consistent with the results of human perceptions.