In this paper a method for adaptive-regularization of image matching 
task is presented. It is based on proposed mixture probability model 
that describes disparity function smoothing and spatially varying 
depending on local intensity gradient.