针对联合双边滤波修复深度图像时无法准确估计滤波邻域范围和权重参数、深度图像的空洞填充效果不佳等问题,提出一种自适应深度图像空洞填充与优化算法。该算法减少了输入参数,实现了对每个深度缺失值的修复,根据有效像素占比确定每个空洞像素点的滤波邻域范围,通过邻域大小计算空间距离权重项参数,引入结构相似度(SSIM)作为颜色相似权重项参数的计算指标。在Middlebury立体匹配数据集和经过配准的Kinect RGB-D数据集上检测所提算法的性能,并与其他方法进行定性比较和定量分析。实验结果表明,所提算法能够有效填充深度缺失空洞,抑制深度图像噪声,更加精细、准确地改善深度图像的质量。

When joint bilateral filtering is used to repair depth images, hole-filling effect is poor because the filtering neighborhood range and weight parameter cannot be estimated accurately. To address this problem, we propose an adaptive hole-filling and optimization algorithm for depth images. The proposed algorithm reduces the input parameters and restores each missing depth value. First, the filtering neighborhood range of each hole pixel is determined based on the effective pixel proportion. Then, the parameter value of spatial distance weight is calculated based on the neighborhood size. Finally, the structural similarity is introduced as a parameter calculation index of the color similarity weight. The performance of the proposed algorithm is tested on the Middlebury stereo-matching dataset and the registered Kinect RGB-D dataset, and qualitative comparison and quantitative analysis are performed to compare the performance of the proposed algorithm with those of other methods. The experimental results show that the developed algorithm can effectively fill in missing depth values, reduce the image noise, and improve the quality of depth images meticulously and accurately.