Phased scene change detection in ubiquitous environments

Abstract

In a ubiquitous environment, video transfer is very important. In particular, transferring of selected video clips obtained through scene change detection is more important than transferring an entire video. In this paper, inter-frame difference values are first computed through combining the χ 2 histogram with the color histogram, as well as normalization. Next, key frames for a cluster are determined through distance clustering and K-mean clustering. Lastly, key frames for a group are determined through a likelihood ratio. According to our experiments, the proposed method surpassed other methods in its ability to detect scene changes due to the use of three steps: difference value calculation, clustering, and key frame extraction. © 2009 Springer Berlin Heidelberg.