A single defocused image depth recovery with superpixel segmentation

Most of the depth restoration algorithms are complicated for a single defocused image, the restoration effect is poor on the image edges, complex textures and shadow areas. In this paper, the depth of a scene is recovered by a single image defocus cue, a novel approach of single defocused image depth restoration is proposed based on superpixel segmentation. First, the simple linear iterative cluster (SLIC) is used to divide the original image into several superpixel modules. Then, the defocus blur amount of each superpixel module is obtained according to the Gaussian–Cauchy mixed model, so as to obtain the superpixel-level sparse depth map. Finally, the cellular automata model is used to optimize the obtained sparse depth map and to restore the true and accurate panoramic depth map. Compared with other methods, the algorithm not only minimizes the error, but also simplifies the process of extending the edge defocus blur to the global. The experimental results on real data show that the method is not only less time-consuming, but also can effectively improve the depth recovery effect in areas with unclear edges, complex textures and shadows. These demonstrate the effectiveness of the method in providing reliable estimates of scene depth.