Disparity estimation by searching parallel and orthogonal to the epipolar lines

In this paper we present a new disparity estimation algorithm. The algorithm combines the low computational load of Viterbi algorithm approaches with field consistency in all directions found in Markov Random Field approaches. The key step in the algorithm is the application of the Viterbi algorithm orthogonal to the epipolar lines in stead of parallel to them. Two disparity fields are obtained, one via the conventional parallel method and one by the new orthogonal method. A combination step then results in a single field with disparity consistency in all directions. Experiments with both natural and synthetic images have been performed. They show that the combined field is consistent in all image directions and that the disparity quality (PSNR) is significantly better than with current parallel methods.