RGBD Image Matting Using Depth Assisted Active Contours
Ji Liu, Wenliang Zeng, Bowen Yang · 2018
This paper proposes a depth assisted active contours for RGBD image matting. The method adopts the confidence map to decide the weights of color and depth images, then uses the weights in order to guide the depth assisted active contours to segment the given image, which results in a good trimap in the process of trimap generation. Take the color image, depth image, and trimap as input, our method extends the alpha matting to estimate the color and depth mattes. Besides, the method provides an appropriate way to combine the two mattes. The experiments with complex natural images demonstrate that our RGBD matting approach is able to generate good matting results.