Contour extraction improved by gradient saliency detection

The problem of accurately extraction of contour from cluttered natural scenes still remain to be solved. As traditional segmentation based on gradient map was facing a bottleneck in boundary detecting, according to the visual diversity of contour and texture in gradient map, we introduced visual attention and implemented saliency detection to enhance contour and inhibit texture. Experiment results showed that edge segmentation and detection based on gradient saliency map improved the performance of contour extraction, at the same time improved the robustness to parameter variation.