Fast edge-preserving depth image upsampler

In this paper, a new image upsampler is proposed to increase depth image resolution fast while preserving edge information. The proposed upsampler is based on a common edge region of color and depth images. In particular, if a vacant pixel in a higher resolution image grid belongs to the common edge region, it is assigned by a pixel selected from five candidates in a local window. A candidate is chosen by minimum cost evaluation; cost is computed by spatial, color, and range weighting functions. Otherwise, the vacant pixel is replaced with a pixel estimated using bilinear interpolation to speed up the process. In terms of a trade-off between depth image quality and computational complexity, experimental results show that the proposed upsampler outperforms other conventional methods, such as the bilinear interpolator and joint bilateral upsampler.