Compression artifacts of the depth images generated by multi-view video coding always lead to seriously geometry distortions in synthesized views.To remove the noise caused by encoding in depth image, a depth image denoising algorithm based on graph cut was proposed.The local binary patterns operator of depth image and corresponding color image were incorporated into this algorithm, and the edge imformation of depth image was effectively kept.Experimental results show that the quality of the virtual viewpoint rendered using denoised depth image is also greatly improved, the proposed approach offers 0.35 dB average PSNR gains comparing with the traditional filtering method, as well as significant subjective improvement in synthesized views.%多视点视频编码给深度图带来的编码块效应,导致合成的虚拟视点中出现严重的几何失真,为去除深度图中编码带来的噪声,提出一种基于图割的深度图去噪算法.将深度图及相应的彩色图的局部二值模式(LBP)算子作为去噪的约束,融入算法中,有效保持深度图的边缘信息.实验结果表明,由去噪后深度图绘制的虚拟视点的质量得到大幅提高,相对目前优秀的滤波方法能平均获得0.35 dB的PSNR增益,主观效果的提升非常明显.
展开▼
