Regions of interest detection with edge and depth information

Traditional saliency map based regions of interest (ROI) detection often has the problems of not able to locate ROI accurately as human visual attention should be, and its ROI shifting usually exists similar problems. In this paper, two kinds of visual features which are edge and depth information are introduced during feature extraction to access more fitable saliency map to human vision perception, and the dilation operation is used to the binary saliency map to connect parts of one object together. Experimental results show that the modified method is more similar to human vision perception, and works better.