Key-frame-based depth propagation algorithm considering Z-axis movement in 2D-to-3D conversion

Abstract

This study proposes a new key-frame-based depth propagation algorithm required for semi-automatic 2D-to-3D conversion of image sequence. The proposed algorithm is a method to perform depth propagation even in image sequence including z-axis movement as precisely as possible. This method performs a bilateral filtering of the depth based on color information and uses a block-based matching at all locations to estimate the motions between frames and compensate them. Here, the degree of movement to z-axis is inferred through homography matrix calculated by matching feature points and the depth map that compensated this was made to be created. It was found from the experiment that depth map could be estimated exactly by reflecting the motion information between frames while preserving the shape of the depth granted by an operator at a key frame. © 2014 International Information Institute.