(An) efficient representation method of disparity map for view interpolation and stereo image compression

Abstract

An efficient representation method of disparity map is proposed, which is optimized for high quality view interpolation and stereo image compression. The representation method is based on image gradient and disparity of stereo images. To obtain high image quality of the view interpolated images from the stereo images, we formulate the error introduced from view interpolation using stereo images and disparity map. In consideration of the derived error function, a new representation method is suggested by minimizing the error under restriction of given number of nodes. First, the disparity map is estimated using the simple block matching method, and the reliable region on which nodes will be placed is obtained using cross correspondence of the disparity values and matching error. Initial nodes are then placed according to the derived error function. Second, successive approximation is conducted by inserting a node per iteration followed by Delaunay triangulation until the number of nodes reaches the target number. The disparity map and error function are recalculated only for the changed triangles after inserting a node. Third, erroneous node points that may cause artifacts in the interpolated images are detected and removed. In addition, special manipulation on the boundary region is performed by inserting additional nodes. Finally, view interpolation is performed.