Single depth map super-resolution: A self-structured sparsity representation with non-local total variation technique (original) (raw)

2017 Eighth International Conference on Intelligent Computing and Information Systems (ICICIS), 2017

Abstract

Recently, depth maps introduce a very effective representation for solving many fundamental computer vision problems. However, modern 3D scanning devices, such as TOF (Time Of Flight) cameras and Microsoft Kinect sensor, provide a huge unbalance between the resolution of the intensity image and its corresponding depth map. Here, we address the problem of single depth map up-sampling using a new non-local total variation decomposition process with a self-learning structured sparsity model. This technique considers the fact of the decomposed components should be regularized by different constraints, hence better representation for depth maps in sparse domain can be achieved. Using different datasets, experimental results demonstrate superior effectiveness in terms of qualitative and quantitative measures.

Sherif Kishk hasn't uploaded this paper.

Let Sherif know you want this paper to be uploaded.

Ask for this paper to be uploaded.