Fast Ray-Space Interpolation Based on Occlusion Analysis and Feature Points Detection

Abstract

A fast algorithm based on depth discontinuity and occlusion analysis is presented for ray-space interpolation. Feature points at depth discontinuity regions in the top and bottom line of sparse epipolar plane image (EPI) is first extracted, and their directions are determined by multiple epipolar matching method. Each of the feature points grows in the nearby epipolar lines until the direction of the grown feature point departures from the direction of the initial point, which means that occlusion occurs or the feature point is in fact a noise. Finally, directional interpolation is implemented within the regions segmented by the feature points. Experimental results show that the new method runs much faster than the pixel matching based and block matching based interpolation methods, additionally, the quality of interpolated EPIs and the rendered arbitrary viewpoint images obtained by the new method is also much better than that of the other two methods.