基于可伸缩备选区域的CVBF算法

CVBF Algorithm Based on Scalable Candidate Region

采用HEVC对3D视频的深度序列进行压缩编码时,会使深度图的边缘出现伪影,最终导致合成的新视点出现空洞影响视频观看质量。CFBV算法采用备选像素代替深度边缘不可信像素,是目前较为有效且简单的深度边缘滤波算法。为了获得更为准确的深度数据,提出基于可伸缩区域的备选像素提取算法,来提高CVBF算法的鲁棒性。滤波过程中,部分备选像素的值由可信度高的区域求平均获得。实验结果表明,使用改进的滤波方法后,合成虚拟视点的PSNR较采用CVBF算法最高提高了0.12dB.该滤波方法可有效提高压缩编码后的深度序列边界数据的可靠性,进而提高虚拟视点合成质量。

When the 3D video is coded by HEVC,the compressed depth sequence usually has artifacts along boundaries,and the large distortions appears in the synthesized views,thus directly declining the video quality.CFBV algorithm uses candidate values to replace the unreliable pixels along boundaries,which is an effective and simple depth map filtering algorithm.In order to obtain more accurate depth data,a candidate values extraction algorithm is proposed based on the scalable region,which can improve the robustness of the CVBF algorithm.In the filtering process,the values of some candidate pixels are calculated by the high confidence region.The experimental results show that by using the improved filtering method,the PSNR of synthesized virtual view is improved 0.12 dB compared with the traditional CVBF algorithm.The proposed method can effectively improve the reliability of the compressed depth map's boundary,and then improve the quality of synthesized virtual views.