Saliency Detection and Tracking Of Stereoscopic Images and Videos

Saliency detection is believed to be an important precursor for a vast number of multimedia processing applications. Besides 2D saliency detection methods, depth feature is also considered to detect saliency in stereoscopic images as well as for videos. This paper presents an enhanced saliency detection framework, depending on the feature contrast of luminance, color, texture and depth, which in turn are extracted from discrete cosine transform coefficients for the purpose of feature contrast calculation. Here, in order to consider the calculation of local and global contrast, a Gaussian model of spatial distance between image patches is adopted. Human vision system actively seeks salient regions and movements in video sequences to reduce the search effort. The motion saliency model detects the moving objects whose motion is salient to its background. In this paper, we propose a novel stereoscopic images and video saliency detection model for detecting the attended regions that correspond to both interesting objects and dominant motions in video sequences. And the salient region on the video is tracked. Keywords—stereoscopic images; video tracking; 3D images.