基于Kinect相机的深度图像空洞修复及超像素分割算法

针对Kinect相机原始深度图像存在空洞的问题,提出了一种结合彩色图像局部边缘信息的深度图像空洞修复算法。首先,通过双边滤波修复较小空洞;其次,根据彩色图像局部边缘信息将较大空洞分为无边缘和有边缘2类;最后,对第1类无边缘空洞进行均值填充修复,对第2类有边缘空洞先根据彩色图像局部边缘特征分割空洞,再分别由外而内逐步修复,从而完成所有的空洞修复。空洞修复完成后,融合深度信息重新建立了线性谱聚类核函数,并基于此提出一种融合深度信息的线性谱聚类超像素分割算法(LSC-D)。实验结果表明,与其他方法相比,提出的深度图像空洞修复算法具有更高的修复准确度,提出的LSC-D超像素分割算法具有更低的欠分割错误率和更高的边界召回率。

A deep image hole repairing method and a superpixel segmentation algorithm based on Kinect camera

Aiming at the problem that there is a hole in the original depth image of Kinect camera, a deep image hole repairing algorithm combining local edge information of color image is proposed. Firs- tly, the smaller holes are repaired by bilateral filtering. Secondly, according to the local edge information of the color image, the larger holes are divided into two types: edgeless and edged. Finally, the average type fill-in repairing is performed on the first type of edgeless holes, and the second type of ed- ged holes are segmented based on the local edge features of the color image and then gradually repaired from the outside to the end, so as to complete all hole repairing. After the hole repairing is completed, the fusion depth is fused to re-establish the linear spectral clustering kernel function. Based on this, a linear spectral clustering superpixel segmentation algorithm (LSC-D) is proposed. The experimental results show that, compared with other methods, the proposed deep image hole repairing algorithm has higher repairing accuracy, and the proposed superpixel segmentation algorithm has lower under-segmentation error rate and higher boundary recall rate.