The Combination of SSE and Tensor Voting for Salient Points Detection in Natural Images

Salient points detection is a bridge which connects the low-level and mid-level visual processing, and is an important cue for contour extraction and perceptual organization. It is an old but challenge work to detect the salient points in natural images in computer vision field. This paper proposes a novel method for saliency detection in natural images, which combines the SSE (Scale Space Edge) algorithm and tensor voting method. We use the result of SSE to be the input of 2-D tensor voting to find the salient edge points of the foreground object in natural image with superfluous or complex background, which is different from the previous work of tensor voting. The results of our experiments in natural images from Berkeley dataset is improved a lot which are compared with the results of the combination of Canny edge detection and tensor voting at different scales for salient points detection in natural images. Meanwhile, the efficiency of tensor voting is also enhanced.