Color Image Segmentation Based On Contrast And GrabCut

Color image segmentation is the basis of image recognition,and it can directly affect the accuracy of recognition.Color image is difficult to be segmented due to its complex background and rich target color.In this thesis,the color image segmentation algorithm is improved by using visual saliency and GrabCut.The main work is as follows.1.A single-target color image segmentation algorithm based on improved convex hull and color contrast was proposed.For using the saliency detection algorithm based on feature contrast will be separate the high-contrast background area from the target.The above problems can be solved by computing the center saliency map according to the improved convex hull.Firstly,the saliency map was calculated based on color contrast with uniqueness and spatial distribution.Next,the convex hull was improved through FH algorithm.Then the center saliency map can be achieved.Finally,the final saliency map was obtained by fusion of the contrast map and the center saliency map.The object was segmented out using Otsu method.Compared with other algorithms,the proposed algorithm is better in accuracy and visualization.2.A multi-target color image segmentation algorithm based on saliency and GrabCut was proposed.When dealing with multi-target images,some background parts maybe appear in the saliency part when using Gestalt principle to compute the center saliency.To solve this problem,the saliency map in the convex hull was firstly calculated based on boundary prior.Then the comprehensive saliency map was obtained by fusion of basic saliency map and the central saliency may.The final objects were segmented out using Otsu method.According to the result of segmentation,the GMM parameters of GrabCut algorithm are initialized.Compared with the traditional GrabCut algorithm,the proposed algorithm has a significant improvement in the segmentation speed and effect.