Stereo Matching by Discontinuity-Preserving Regularization
Du-Yeong Jung, Juhyun Oh, Sang-Chan Lee, Chul-Hun Lee, Ki-Gon Nam · Journal of Electrical Engineering and Information Science · 1999
To extract 3D information, such as disparity and depth from stereo images, is mathematically an ill-posed problem. To overcome this problem, and to obtain a disparity map, regularization is applied. The addition of the smoothness assumption to solve this problem leads to an undesirable blurring of the disparity map across discontinuities. The discontinuities in the disparity map can be preserved by using a controlled-continuity stabilizer if the discontinuity information is known in advance, which is generally not possible. In order to obtain the discontinuity information, this paper focuses on the fact that the disparity map obtained by regularization has high-energy distribution at the discontinuities. Once the disparity map is obtained, it is used as the initial condition for the discontinuity information needed for the next discontinuity-preserving regularization. The algorithm has been tested on several stereo images and the results have shown that the blurring of disparities across discontinuities is drastically reduced.