Automatic Multi-resolution Joint Image Smoothing for Depth Map Refinement

In this paper, we present a technique to remove the noise of a depth map while fill the missing regions of the depth map. Generally, the depth map is degraded during the sensing process, thermal noise, the condition of the atmosphere, and the occlusion by the objects. Different to the previous works which only adopt the joint image filters directly, we propose an automatic multi-resolution approach with a probabilistic Bayesian model to remove the noise of the depth map while fill the missing regions. Our model is based on the joint guided filtering and cascaded with a messing-passing technique called belief propagation. As compared to the state-of-art joint image filtering and image smoothing, the experimental results demonstrate that our proposed approach is promising.