A novel 2D-to-3D conversion method based on blocks world

The technique of 2D to 3D conversion consists of two main steps: depth estimation and virtual view rendering. The depth estimation from a single monocular image is the key challenge for 2D to 3D conversion. We propose a novel depth estimation method based on Abhinav G. etc's work “blocks world” and then the other virtual view is synthesized with DIBR (Depth Image Based Rendering). The input image is first segmented into several “blocks” with different depth and supporting relationships using Abhinav G.'s method. Assuming that the scene contains only ground and sky, the gradient depth can be assigned with the horizon (the contact line between ground and sky) being the farthest depth. Then the depth of “blocks” in the scene will be determined by the contact line between the block and the ground. Through joint bilateral filter, the final depth map of good quality is obtained. Experimental results show that our depth map is far more accurate than state-of-art algorithms and resulting 3D images are of good 3D perception.