The goal of stereo-matching is to estimate the depth of a scene from the disparity between the left and right stereo images. The state-of-the-art algorithms use global optimization to estimate the disparity, using some efficient combinatorial search. These global algorithms, however, tend to be rather slow when both the size of the stereo images and the range of disparities are large. This can well be appreciated from the fact that almost all real-time algorithms (even for moderately-sized images) are based on greedy strategies (local optimization without any global reasoning).

The main motivation of this project is to develop a real-time algorithm for estimating the disparity (depth profile) of a biological specimen from the left and right images captured by a stereo-microscope. The goal is to develop a simple and fast algorithm by combining the speed of greedy strategies with the processing capabilities of an adaptive form of the bilateral filter. The role of the filter is to reduce the image ambiguity and enhance the discriminative power for the correspondence search. We have sufficient evidence to believe that this could give promising results.

During the project, the student will investigate some existing ideas in the literature, and test some original ones as well. We will provide the student with an efficient Java implementation of the bilateral filter.