Depth image super-resolution is an extremely challenging task due to theinformation loss in sub-sampling. Deep convolutional neural network have beenwidely applied to color image super-resolution. Quite surprisingly, thissuccess has not been matched to depth super-resolution. This is mainly due tothe inherent difference between color and depth images. In this paper, webridge up the gap and extend the success of deep convolutional neural networkto depth super-resolution. The proposed deep depth super-resolution methodlearns the mapping from a low-resolution depth image to a high resolution onein an end-to-end style. Furthermore, to better regularize the learned depthmap, we propose to exploit the depth field statistics and the local correlationbetween depth image and color image. These priors are integrated in an energyminimization formulation, where the deep neural network learns the unary term,the depth field statistics works as global model constraint and the color-depthcorrelation is utilized to enforce the local structure in depth images.Extensive experiments on various depth super-resolution benchmark datasets showthat our method outperforms the state-of-the-art depth image super-resolutionmethods with a margin.
展开▼