A VIEW INTERPOLATION METHOD WITHOUT EXPLICIT DISPARITY ESTIMATION

In this paper, we present a new view interpolation method based on the Nonlocal-Means (NL-means). Using NL-means, every pixel in the intermediate view is set to a weighted linear combination of the averages of pairs of pixels in the reference views. The pixels of each pair are in the same vertical position and in the symmetric horizontal positions from the viewpoint of the pixel being interpolated. The weights of the linear combination are calculated based on intensity closeness between the neighborhoods of two pixels in each pair. The main benefit of this method is that it does not require explicit disparity estimation. Experimental results show that the proposed method outperforms other view interpolation algorithms.