摘要
为解决Kinect感应器所采集的深度图像中存在大面积空洞的问题,提出了一种深度图像空洞修复方法。该算法首先输入同步获取的彩色图像和深度图像;接着利用k-means算法对灰度化后的彩色图像进行聚类,聚类结果作为引导图像;然后对每个深度图像中的空洞点,搜索引导图像中与之相匹配的非空洞像素点,将该点的深度值作为空洞点的深度值。实验结果表明,该算法利用聚类思想,将彩色图像应用到对深度图像的空洞修复,有效完成了对深度图像的空洞填充,修复后深度图像的平滑度优于联合双边滤波方法,较好地提高了深度图像的质量。
In order to solve large dark holes in Kinect depth image, this paper proposes a depth hole- filling method. It firstly inputs synchronous color image and depth image, and uses k - means algorithm to cluster image pixels in gray image. The result is used as a guiding image. Then, for each hole of the depth image, it finds a non - hole pixel matched in the guiding image and uses its depth value to fill the corresponding hole. The experimental results show that the proposed algorithm, using clustering concept, applies the color image to the hole repairing of depth image and effectively fills dark holes in the depth image, and as the smoothness of repaired depth image is better than that of joint bilateral filtering method, the quality of depth image improves a lot.
出处
《微处理机》
2015年第4期42-44,48,共4页
Microprocessors
基金
国家自然科学基金(61101158
61471157)
江苏省自然科学基金(BK20141159)