A 2D to 3D conversion method based on support vector machine and image classification

Abstract

With the development of 3D technology, converting 2D videos available into 3D videos has been an important way to gain 3D contents. In the conversion, a crucial step is how to obtain a more accurate depth map. This paper proposes a method for depth extraction based on color and geometric information of the original image. Firstly, we generate a qualitative depth map by SVM and classify image scenes into three categories. Then depending on geometric information, a geometric depth map can be generated by vanishing lines detection and gradient plane assignment. At last, we blend two depth maps to get a final depth map, which has more widely application and improves accuracy of depth better.