Comparison of Depth Super-Resolution Methods for 2D/3D Images

For a few years cameras have been available which are able to provide depth values gained from a PMD chip through the Time-of-Flight principle. Recently, cameras combining a normal color chip with such a PMD chip in a monocular setup have been developed. One drawback of these 2D/3D cameras is that the resolution of the depth images is much lower than those of the color images due to the limited resolution of current PMD chips. This holds true whenever PMD cameras and normal cameras are used together. However, for certain applications high resolution depth images are desirable. The color images can be utilized to generate high resolution depth images which are closer to the ground truth than the depth images produced with common scaling methods. A widely spread method to fuse the color and depth images is cross bilateral filtering and this rather general method is adopted in several approaches. In this paper different bilateral filtering strategies are compared in theory as well as in practice and especially the iterative application is addressed. Similar approaches based on a cost volume or on Markov Random Fields are addressed additionally.