Fast Stereo Matching Using a Bi-Exponential Edge Preserving Smoother

Oliver Akermann

Semester Master Project

Microengineering section, EPFL

June 2011

Abstract

The goal of this semester project was to develop a fast stereo matching algorithm and to implement it as an ImageJ plug-in. The developed algorithm is based on the Adaptive Support-Weights (ASW) approach described in [1]. The ASW method is very similar to a bilateral filter. In this project, the bilateral filter was replaced by an adaptation of the Bi-Exponential Edge-Preserving Smoother (BEEPS), which is a filter yielding similar results to the bilateral filter while being much faster.
The results showed that the implemented algorithm is up to 150 times faster than the ASW approach. This acceleration comes at a cost: The quality of the resulting disparity maps is somewhat lower. However, the quality is still comparable to other stereo matching methods.
[1] K. Yoon and I. Kweon, "Locally Adaptive Support-Weight Approach for Visual Correspondence Search", IEEE Conference on Computer Vision and Pattern Recognition, vol. 2, pp. 924-931, 2005.

Fig. 1: Left image of a stereo image pair

Fig. 2: Disparity map obtained with BEEPS-based method