Depth-map Super-resolution Algorithm with Redescending M-estimator Constraints

Most of the depth-map super-resolution algorithms rely on the information provided by the guided color image. Due to differences in structure between guided and input signals, such algorithms are hard to preserve depth boundaries. We address this problem by redescending m-estimators. First, the neighboring constraints for depth are built based on color similarities. Second, redescending m-estimator is used to measure the constraints. Then, the depth super-resolution is formulated as an optimization problem. Such a choice helps in dealing with violations of the assumption that similar colors have similar depth. Finally, the solution is obtained by the generalized iteratively reweighted least squares. The experimental results demonstrate that our algorithm can preserve depth boundaries and is superior to existing algorithms in terms of depth accuracy.