Single-Image Depth Inference Based on Blur Cues

With the rapid advancement of 3D visual technology, the technique of depth inference from a single image has received new attention. In this work, we present several single-image depth inference algorithms based on the blur degree in different regions in one image. We identify two major sources of image blur: camera defocus and atmospheric reflectance. The latter is also known as the haziness. We build models for these two scenarios with the depth information as a model parameter. Thus, we are able to infer the depth information from the observed image. Experimental results are conducted on a large variety of images to demonstrate that the robustness of the proposed depth inference method.