一种深度图像修复算法研究

摘要

For depth images of Kinect sensor with large amounts of noises and holes,an improved depth image inpainting algorithm based on region segmentation and joint bilateral filtering is presented.The noise characteristics of depth image tend to vary in different regions,and the existing depth image inpainting algorithm suffers difficulties in an appropriate selection of parameter due to the disturbance of noise,which poses obstacle to the filtering process and degrades the preservation of depth image structures.Therefore,noise characteristics within regions are utilized to derive parameters corresponding to the involved pixels.The experimental results show that the joint bilateral filtering algorithm based on region segmentation can efficiently remove the noise and filling the holes of depth image,and can preserve the structure information of depth image as well.%针对Kinect传感器获取的深度图像包含大量噪声和缺失深度信息的孔洞等缺点,提出一种基于区域分割和联合双边滤波的深度图像修复算法.深度图像中不同区域具有不同的噪声特性,现有的深度图像修复算法采用单一的滤波参数对整幅深度图像进行处理具有盲目性和无法有效保持滤波后深度图像的结构信息的缺点.文中在深度图像分割的基础上,根据分割区域内的噪声属性自适应地确定该区域滤波参数,实现深度图像的去噪和孔洞修复.文中算法与联合双边滤波算法对Kinect深度图像修复的对比实验表明,文中算法能有效去除深度图像的噪声和修补缺失深度信息的孔洞,且修复后的深度图像峰值信噪比高于联合双边滤波算法处理后的深度图像峰值信噪比,并能较好的保护深度图像的区域结构.