Depth map restoration based on Kinect camera

Due to the instability of depth data, image noise, lack of depth information and other reasons, the quality of Kinect depth maps can easily be reduced. In this paper, a color-guided depth map restoration optimization model is proposed to address the issues mentioned in a unified framework based on the structure information from color and input depth. By adopting a robust non-convex function and multi-channel weight function to simulate the regularization term of optimization model, this method can be used to suppress texture copy artifacts caused by the inconsistency between depth discontinuities and color fringes. Then, a modified fast alternating direction method of multipliers is used to solve the optimization problem. Through comprehensive experiments on both simulated data and real data, the promising performance of the method is proved. Moreover, the results show that the proposed scheme is around two times faster than the conventional majorize-minimize solver method.