Depth Map Super-Resolution Reconstruction Based on Second-order Total Generalized Variation Guided by Color Segmentation
Meifu Zhao, Yu Wang · 2023
In the context of the depth map super-resolution reconstruction algorithm guided by a color image, there is a problem of inconsistency in edge structures between the color image and the depth map, resulting in texture copying artifacts. To address this issue, this paper proposes a depth map super-resolution reconstruction method based on second-order total generalized variation guided by color segmentation. Firstly, the low-resolution depth map is initially interpolated, and the resulting image is divided into edge regions and smooth regions. Next, the color image is segmented using the improved simple linear iterative clustering (SLIC) algorithm combined with the standard deviation-based local binary pattern (LBP) descriptor. The segmentation result that aligns with the depth edge structures is exported. Subsequently, the joint trilateral filtering algorithm based on the segmentation result of the color image is used to interpolate the pixels in the edge regions of the depth map, which helps to avoid interference from texture information. Finally, the up-sampled depth map is input into the second-order total generalized variation (TGV) model combined with multi-scale morphological gradient, to reconstruct clearer edges by incorporating more accurate gradient information. Experimental results demonstrate that our method effectively suppresses texture copying artifacts and preserves edges well.