深度图辅助的主动轮廓分割算法 postprint

Abstract: The similarity of foreground and background is a tough question in image segmentation. Based on the active contours, combining with the depth map, this paper proposes a depth-assisted region-based active contour method. It uses the filtering algorithm to recover the depth map firstly, then calculates the confidence maps of the RGB image and the depth image by Gaussian Mixture Model, finally uses the confidence map to calculate the weights of the colors and depths in the given region to guide the segmentation process. The proposed algorithm uses the local information of color and depth images to guide the segmentation, which obtain a more accurate segmentation. The experimental results show that the proposed method obtains the segmentation results are nearer to the true circumstance, also it improves the accuracy of image segmentation.