Single image depth estimation based on dark channel prior

This paper presents a depth estimation method based on dark channel prior to construct a depth map from a single 2D image of an outdoor scene for 3D display applications. The method computes a depth map using a series of image processing strategies such as dark channel computation, edge detection, the edge-based iterative thresholding technique, the region split-and-merge algorithm, the foreground and background classification, and the depth estimation by integration of dark channel and elevation information. Experimental results and comparisons show that the proposed method provides good depth estimation in boundaries and regions of sky and buildings.