Video annotation based on temporally consistent Gaussian random field

A novel method for automatically annotating video semantics, called temporally consistent Gaussian random field (TCGRF) is proposed. Since the temporally adjacent video segments (e.g. shots) usually have a similar semantic concept, TCGRF adapts the temporal consistency property of video data into graph-based semi-supervised learning to improve the annotation results. Experiments conducted on the TRECVID data set have demonstrated its effectiveness