Abstract:The edge information plays a key role in the restoration of a depth map. Most conventional methods assume that the color image and depth map are consistent in edge areas. However, complex texture regions in the color image do not match exactly with edges in the depth map. In this paper, firstly, we point out that in most cases the consistency between normal map and depth map is much higher than that between RGB-D pairs. Then we propose a dual normal-depth regularization term to guide the restoration of depth map, which constrains the edge consistency between normal map and depth map back and forth. Moreover, considering the bimodal characteristic of weight distribution that exists in depth discontinuous areas, a reweighted graph Laplacian regularizer is proposed to promote this bimodal characteristic. And this regularization is incorporated into a unified optimization framework to effectively protect the piece-wise smoothness(PWS) characteristics of depth map. By treating depth image as graph signal, the weight between two nodes is adapted according to its content. The proposed method is tested for both noise-free and noisy cases, and is compared against the state-of-the-art methods on both synthesis and real captured datasets. Extensive experimental results demonstrate the superior performance of our method compared with most state-of-the-art works in terms of both objective and subjective quality evaluations. Specifically, our method is more effective on edge areas and more robust to noises.

Reconstructing Sparsely Represented High-resolution Depth Map from Corrupted Samples

Depth Map Inpainting Using a Fully Convolutional Network

Depth map enhancement based on color and depth consistency

Multi-View Depth Map Sampling for 3D Reconstruction of Natural Scene

Consistent depth maps recovery from a video sequence.

Data Driven Robust Image Guided Depth Map Restoration

Recovering Consistent Video Depth Maps Via Bundle Optimization

CRF-Based Reconstruction from Narrow-Baseline Image Sequences.

Depth Upsampling Method Via Markov Random Fields Without Edge-Misaligned Artifacts

Depth Map Completion by Jointly Exploiting Blurry Color Images and Sparse Depth Maps

Spatio-Temporal Depth Recovery of Dynamic Scenes with Multiple Handheld Cameras

Consistent Depth Maps Recovery from a Trinocular Video Sequence.

Depth Map Upsampling via Progressive Manner Based on Probability Maximization.

Depth Map Super-resolution Based on Dual Normal-depth Regularization and Graph Laplacian Prior

Depth Map Super-Resolution Based on Joint Dictionary Learning

Guided Depth Map Super-Resolution Using Recumbent Y Network

Depth Restoration: A fast low-rank matrix completion via dual-graph regularization

Depth Map Denoising Using Graph-Based Transform and Group Sparsity

Depth Completion using Piecewise Planar Model

Depth Completion Towards Different Sensor Configurations Via Relative Depth Map Estimation and Scale Recovery