A stochastic stereo matching technique based on beta-splines

The authors cast the matching problem as a stochastic optimization problem with the disparities modeled by beta-splines. They regard the positions of the control vertices as parameters that are evaluated by an optimization algorithm. The process is ultimately the determination of the lowest-cost state whose resulting parameters defines the registration map. To this end, they introduce a cost function that is insensitive to a possible bias or scale between the intensities of the two images The algorithm is tested with both simulated as well as real radar data and issues pertaining to its efficiency are discussed.