Color Image Segmentation based on Color Contrast and Improved Convex Hull

Abstract: Since that it can enhance the Region Of Interest (ROI), saliency detection is important for imaging processing. Most saliency detection approaches are based on feature contrast to find the salient area. However, for the similar contrast between background and foreground, the traditional approaches always fail to differentiate these regions of the background from the foreground. To solve the similar contrast problem, this paper presents a novel method for color image segmentation based on improved convex hull and color contrast. Firstly, the method takes superpixel as the basic processing unit, and measures the contrast saliency map based on color contrast with uniqueness and spatial distribution. Secondly, it calculates convex hull using color boosted Harris corner points which was improved by FH algorithm. Based on the improved convex hull, the center saliency map is obtained. The final saliency map is constructed using the contrast saliency map and the center saliency map. Finally, the object is segmented out from the final saliency map using Otsu method. Compared with several state-of-the-art methods on MSRA 1000 and ECSSD dataset, the proposed method achieves better performances in visualization, precision and recall.

Keywords: Superpixel, color contrast, improved convex hull, saliency map, color image segmentation.