Automatic object-based 2D-to-3D conversion

Abstract

This paper presents a new 2D-to-3D conversion method based on structure from motion (SfM) in order to relieve the visual fatigue and simultaneously improve the three-dimensional effect (3D effect). First, we obtain the 3D information such as camera positions and depth values via SfM. Then, we segment an input image, and find the nearest object region. Next, the projective matrix of the nearest object is computed. Finally, the nearest object is warped using the computed projective matrix, and the other regions are properly warped according to their depth values. Experimental results show that the proposed method can synthesize stereo views better than the other methods.