Spatially Adaptive Tensor Total Variation-Tikhonov Model for Depth Image Super Resolution

Depth images play an important role in 3-D applications. However, due to the limitation of depth acquisition equipment, the acquired depth images are usually in limited resolution. In this paper, a spatially adaptive tensor total variation-Tikhonov model is proposed to solve this problem. The tensor total variation regularization is adopted to maintain sharp edges that reflect latent discontinuities in the real world, while the Tikhonov regularization ensures that depth changes smoothly inside objects. Furthermore, a fused edge map is proposed to indicate edge regions and balance both regularization terms. In edge regions, tensor total variation regularization is predominant, thus edge blurring artifacts are suppressed. In non-edge regions, Tikhonov regularization plays a more important role to suppress staircasing artifacts. Specifically, texture edges are removed in the fused edge map, and texture copying artifacts are avoided. Experimental results demonstrate the effectiveness and superiority of the proposed framework. Moreover, the proposed method yields much sharper edges and a lower percentage of bad pixels.