PCA-based approach for video scene change detection on compressed video

Abstract

An automatic, real-time detection approach to video scene change detection is presented. Owing to the high correlation of two consecutive video frames, it is proposed that only the eigenvector corresponding to the largest eigenvalue is retained in the principal component analysis (PCA) for video data. A one-dimensional PCA feature of video data is then generated from the PCA. It shows superior performance compared to the histogram feature and the pixel feature. The detection algorithm based on this PCA feature is then designed to detect both abrupt and gradual transitions. The proposed approach is tested on the TREC video test repository to validate its performance. © 2006 The Institution of Engineering and Technology

PCA-based hierarchical approach for video gradual transition detection in compressed domain

“…In our previous work [14], we found this PCA-based feature is more suitable to be used as the input of the SBD than other features in the C-domain. In our recent work, we have analyzed and prove its superior performance through the theory approach.…”

“…In this paper, for the video frames, we also analyze the basis image constructed by each eigenvector of the training set. As shown in Fig.1, the basis images correspond to those eigenvalues in the Table 1 in [14]. It can be observed that the basis image of the largest eigenvalue usually reflects the 'unchangeable' part or common information of the training set.…”

“…Lastly, for the decision rules, in the early work [1][2][3][4]14] global threshold is predefined and used. But the SB is a local activity in the temporal domain, and it is a very hard work to choose a suitable threshold for different videos.…”

“…Given the above eigenspace is derived from the arbitrary one of two adjacent shots. When the DC vector k X of each frame [14] in these two shots projects into this new eigenspace, the projection feature has following characters: firstly, within the same shot, due to high visual content coherence, the projection features of image series have smooth fluctuations; Secondly, since different shots have general-different contents, the projection features change dramatically from one shot to another shot, as shown in fig.1 in [14]. Therefore, the feature can be a trade-off between invariance within a shot and sensitivity to the shot transition.…”

“…However, the GT is more difficult to be detected and remains a challenging problem by far [1][2][3][4][5]. In our previous work [14], we proposed a PCA-based approach for the SBD in the compressed domain. We found the superior performance of the PCA-based feature.…”