Spatial-temporal salient unit detection based on features in video sequences

A spatial-temporal salient unit detection method is proposed to express the video content. In the process of extracting the salient regions in spatial domain, brightness and direction information are used to construct the synthetically salient image. Threshold method is selected to segmentation. By morphological processing and area filtration, saliency degree is calculated. And the salient regions are determined under competition mechanism. In the process of extracting the salient regions in temporal domain, by definition a merger criterion, the high relevant regions are merged. Video´s content of a lens can be expressed by all spatial-temporal salient regions. The experiments are performed to demonstrate the validity from spatial and temporal aspects.