基于边缘增强的深度图超分辨率重建

准确的深度图像获取是计算机视觉中的一个难题 。传统的立体匹配得到深度的方法不仅计算量大,而且在纹理稀疏与重复区域往往存在较大 的误差。主动式深度传感器虽然解决了这些问题, 但其获取的深度图存在着分辨率低和易受噪声干扰的问题。因此,本文提出一种结合彩色图 像信息 的深度图超分辨率(SR)重建方法来提高深度图的质量与分辨率。首先运用自回归(AR)模型 下的非局部均值(NLM)算 法获取初始的上采样深度图；然后利用边缘提取与边缘修复算法优化深度图。实验结果表 明,本文提出的方法能够生成误差更小、主观质量更好的高分辨率深度图。

Super-resolution reconstruction for depth map based on edge enhancement

Capturing accurate depth map is a tough task in computer vision.Tradi tional method of stereo matching not only has a large computational cost,but also has a bad performance in areas with sparse or repeated texture.Although active range sensor can solve these problems,the depth maps cap tured by the range sensors have low resolution and are susceptible to noise.In t his paper,a super-resolution reconstruction method based on the edge enhancemen t algorithm is presented to improve the quality and resolution of the depth map. Firstly,nonlocal means (NLM) algorithm is applied in the autoregressive (AR) mod el to generate an up-sampled depth map.Afterward,accurate edge detection and edge rectification algorithms are used to optimize the up-sa mpled depth map.The experimental results indicate that the proposed approach,in general,is capable of generatin g the high-resolution depth maps with smaller error and better subjective quality in comparison with the art-of-stat e algorithms.