Segmentation of Lines and Arcs and its Application for Depth Recovery
ICASSP '97 Proceedings of the 1997 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP '97) -Volume 4 - Volume 4
Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images
IEEE Transactions on Pattern Analysis and Machine Intelligence
Hi-index | 0.00 |
We present a coarse-and-fine algorithm that also uses discontinuity edge elements for stereo matching. The model uses a pyramid of multi-resolution scale images as input to create multiple disparity maps and discontinuity maps for the horizontal and vertical edges. The energy cost model includes the data constraint within each grid layer, smoothness constraint within each grid, and the inter-grid constraint to make sure there is consistency between neighboring resolution grids. Edge elements have an effect on the smoothness constraint in all the resolution levels to reduce an error from trying to smooth between discontinuity regions. We also propose a method to propagate the edge constraint between neighboring resolution grids. Using the Gibbs Sampler with simulated annealing optimization, we obtain the disparity and edge maps as a result of our algorithm. Results for various test images are presented.