融合纹理信息的深度图像修复

针对低质量深度图像中存在的空洞和噪声问题,提出了融合纹理信息的深度图像修复算法。首先利用形态学操作对空洞进行优化,基于区域的分割算法完成空洞区域分割;然后将提取出的空洞区域进行纹理信息填补,分析空洞区域与同场景彩色图像的局部结构相似性,完成空洞初修复;最后利用灰度级图像重建算法,对边缘空洞区域进行填充和平滑处理。基于标准数据集Middlebury,所提算法与快速行进算法、自适应中值滤波算法和形态学重建算法相比,对大面积空洞信息完成了良好的修复。在获得较好修复效果的同时,该算法保持了图像的结构完整性和整体平滑性。

Depth maps inpainting with fused texture information

To solve the problem of holes and noise in low quality depth maps, a depth maps inpainting algorithm based on fused texture information is proposed. Firstly, the cavity is optimized by morphological operation, and the segmentation of the cavity is completed by a region based segmentation algorithm. Secondly, the extracted cavity region is filled with texture information, and the local structural similarity between the cavity region and the same scene color image is analyzed. Finally, the gray scale image reconstruction algorithm is utilized to fill the edge hole region and smooth the region. Compared with the fast marching algorithm, the adaptive median filtering algorithm and the morphological reconstruction algorithm on the standard dataset Middlebury, the proposed algorithm completes the large area cavity information. While getting better inpainting results, the algorithm maintains the structural integrity and overall smoothness of the image.