Abstract:One of the major problems of Kinect depth image is the existence of noise and holes. Thus, it is necessary to enhance depth images for post processing. In this paper, we propose an adaptive joint bilateral/trilateral filter, considering both the depth image and the aligned colour image, to refine the depth edges and remove the noise. Specifically, the pixels around the common boundary of depth edges and colour edges are taken as unreliable pixels due to the perturbations around the depth edges, then a support region and kernel-width adaptive joint bilateral filter considering the colour edge is applied to refine the marked unreliable pixels and the depth edges. For the depth pixels except unreliable pixels, we apply a kernel-width adaptive joint trilateral filter to remove the noise. Experimental results show that our approach outperforms the compared methods in terms of finer depth edges and noise removal.

Kinect Depth Image Enhancement with Adaptive Joint Multi-Lateral Discrete Filters