Abstract: Summarization of video content is a complex task that requires feature selection and frame matching. To extract meaningful frames, named key frames, we have proposed partitioning of frames with Voronoi diagrams for further region matching throughout the video sequence. A unique partition metric has been used that takes into account color and textural, structural and geometric properties of Voronoi regions. Feature set designed for CBIR and CBVR has been analysed. The reasonable feature selection and incorporation into frame matching process has permitted to obtain competitive results. All these actions have allowed revealing significant changes in content while omitting slight deflections and repeats. Key frame extraction procedure has been described in detail. The proposed method has been checked on different test samples and compared with existing methods for precision and recall.

Keywords: key frame extraction, generator point detection, Voronoi diagram, partition metric, feature set.

ACM Classification Keywords: I.2.10 Vision and Scene Understanding (Video analysis), I.4.6 Segmentation (Edge and feature detection)

Link:

KEY FRAME PARTITION MATCHING FOR VIDEO SUMMARIZATION

Olena Mikhnova, Nataliia Vlasenko

http://www.foibg.com/ijima/vol02/ijima02-02-p08.pdf