Spatial-Depth Supre Resolution for Range Images

Qingxiong Yang*, Ruigang Yang*, James Davis+, David Nistér*


Intermediate results. (a)Camera image. (b)The initial depth map. (c)Depth map after one iteration. (d)Depth map after three iterations. (e)Depth map after ten iterations.

Abstract

We present a new post-processing step to enhance the resolution of range images. Using one or two registered and potentially high-resolution color images as reference, we iteratively refine the input low-resolution range image, in terms of both its spatial resolution and depth precision. Evaluation using the Middlebury benchmark shows across-the-board improvement for sub-pixel accuracy. We also demonstrated its effectiveness for spatial resolution enhancement up to 100X with a single reference image.

Experimental result


  • Bilateral filtering with disparity map.

    From left to right: color images, low-resolution disparity maps, refined disparity maps. Note that the original resolutions of the two depth maps are 28x23 and 56x47 respectively, and the resolution of the refined disparity maps is 450x375.

  • Depth enhancement.

    Depth enhancement. Left: synthesized view using depth map produced by DoubleBP stereo algorithm. Right: after depth enhancement.

  • Sub-pixel estimation verification.

    Sub-pixel estimation verification. The scores on the last four columns are the average ranks with error threshold 0.5. The scores with bold font are among the top 10 performers. The entries with blue highlighting are stereo algorithms originally without sub-pixel estimation, the others are algorithms originally having sub-pixel estimation. The scoring scheme is the same as the Middlebury benchmark. We show across-the-board improvement for sub-pixel accuracy.

    Sub-pixel benchmark(2-view, 1-view)

    More experimental results.

    Related Paper

  • Qingxiong Yang, Ruigang Yang, James Davis, David Nistér, Spatial-Depth Super Resolution for Range Images, CVPR 2007 (acceptance rate: 27.5%). [PDF[7.1MB] | PDF[0.6MB]]