Depth estimation from a single image using dark channel prior and color information

This paper presents a depth estimation method from a single outdoor image using dark channel prior and color information. The method computes a depth map from a single outdoor image for 3D display applications using a series of image processing algorithms and techniques such as thresholding technique, color segmentation, edge detection, region split-and-merge. The procedure of proposed method can be categorized into three steps, i.e., image segmentation, foreground and background partition, and depth generation. The dark channel prior, color, and edge information are used to segment images by edge-based Otsu thresholding and mean shift clustering. Then a region split-and-merge algorithm is executed for advanced image segmentation. We have also developed a knowledge-based foreground and background partition algorithm using color and edge information. The depth map is generated from the individual foreground and background depth estimation. Experimental results and comparisons show that the proposed method produces clearer depth maps with distinct regions.