A Weighted Color MRF Model for 3D Reconstruction from a Single Image

Abstract

Image-based modeling problem is a tough and basic problem in computer vision and computer graphics. This paper provides an improved solution to generate a scene model from a single image based on the multi-scale Markov random field. We consider the different influence of the different color components in RGB color space when modeling through a single still image. In this paper, we proposed a weighted color model for 3D reconstruction and draw a conclusion that different color components contain different depth information, and by modifying the weight of different color components in the image we can get a better visual effects and more accurate 3D model. The output model of our solution then can be used to construct a virtual scene rapidly based on a scene image.