3D-HEVC深度图预处理与误码掩盖技术研究

发布时间：2018-01-23 01:07

本文关键词： HEVC 3D-HEVC 深度图预处理误码掩盖　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：多视点视频加深度(MVD)由于具有良好的后向标准兼容性,且利用基于深度图像的绘制(DIBR)技术能够在解码端合成任意视点的图像,实现任意位置视点的观看。运动图像专家组(MPEG)将MVD选作新一代三维视频(3DV)编码标准的数据格式。3DV标准可分别基于H.264/AVC和H.265/HEVC标准进行制定。相比于H.264/AVC,由ITU-T和ISO联合制定的新一代视频编码标准HEVC在相同重建质量的前提下,能够节省一半左右的码率,因此3D-HEVC受到了更多关注。由于引入了深度信息,增加了编码码率,因此提高深度图的编码效率成为了3D-HEVC的关键。另外,码率的增加也使得带宽有限传输网络中出现网络拥塞的风险增加,易导致视频信息丢失,解码错误现象。因此,针对3DV的特点制定更加有效的误码掩盖技术具有重要意义。本文重点研究了用于提高深度编码效率的深度图预处理技术以及3DV误码掩盖技术,主要研究内容如下:由于深度估计算法的限制,使得获取的深度图质量较低。具体表现为深度图中物体的边界信息与视频图像中真实的物体边界之间一致性较差,出现了较多的虚假边缘。虚假边缘不仅额外增加了编码码率,也降低了解码端虚拟合成视点图像的质量。为了去除虚假边缘,提高深度图的编码效率,本文提出了一种深度图预处理算法。所提算法根据深度图与视频图像中物体边缘的相关性,利用各向异性扩散方法对虚假边缘进行了平滑。由于所提平滑方法具有各项异性的特点,在滤除虚假边缘的同时,能够对深度图中与视频图像物体边缘相关性较高的真实边缘信息进行有效保护。利用所提算法可以在不降低虚拟合成视点图像质量的前提下,有效地滤除虚假边缘,提高深度图的编码效率。对于在网络传输过程中丢失的帧信息,一般利用其与前后帧的时域相关性进行插帧实现误码掩盖,但是对于运动较为剧烈的前景部分却难以准确恢复。相比于二维视频,MVD信息的视点间相关性也可以用于丢失帧的插值过程。本文提出一种区分前景背景的插帧技术用于实现3DV的误码掩盖。首先通过对前后帧的帧差进行门限比较,初步划分出前景与背景,并进一步利用形态学滤波的方法对进其进行去噪处理。对于背景部分,仍然利用前后帧进行插值;而对于运动较为剧烈的前景部分,则利用相邻视点在当前视点的虚拟合成图像进行填充。所提方法综合利用了时域和视点相关性,提高了重构帧的质量。
[Abstract]:Multi-view video plus depth MVDs have good backward standard compatibility and can synthesize arbitrary view images at the decoding end by using depth image-based rendering (DIBR) technology. Realization of viewing at any position. Moving Image expert Group (MPEG) selects MVD as a new generation of 3D video player 3DVV). Coding standard data format. 3DV standard can be based on H.264% AVC and H.265% HEVC standard, compared to H.264% AVC. The new generation video coding standard HEVC, which is jointly developed by ITU-T and ISO, can save about half of the bit rate under the same reconstruction quality. So 3D-HEVC attracts more attention. Because of introducing depth information and increasing coding rate, improving the coding efficiency of depth map becomes the key of 3D-HEVC. The increase of bit rate also increases the risk of network congestion in limited bandwidth transmission network, which easily leads to the loss of video information and decoding error. It is of great significance to develop a more effective error concealment technique according to the characteristics of 3DV. This paper focuses on the depth map preprocessing technology and the 3DV error concealment technology which are used to improve the depth coding efficiency. The main research contents are as follows: because of the limitation of depth estimation algorithm. The quality of the obtained depth map is low, which is shown by the poor consistency between the boundary information of the object in the depth map and the real object boundary in the video image. There are more false edges. False edges not only increase the coding rate, but also reduce the quality of the decoded virtual composite view image. In order to remove false edges and improve the coding efficiency of depth map. In this paper, a depth map preprocessing algorithm is proposed, which is based on the correlation between the depth map and the object edge in the video image. The method of anisotropic diffusion is used to smooth the false edges. Because the proposed smoothing method has the characteristics of heterogeneity, the false edges are filtered at the same time. It can effectively protect the real edge information which is highly correlated with the object edge of video image in depth map. The proposed algorithm can not reduce the quality of virtual synthetic view image. The false edges are filtered effectively and the coding efficiency of depth map is improved. For the frame information lost in the process of network transmission, the time domain correlation between the frame and the front and rear frames is generally used to implement error concealment. However, it is difficult to recover accurately the foreground part of the motion which is more intense. Compared with two-dimensional video, it is difficult to recover accurately. The interview correlation of MVD information can also be used in the interpolation process of lost frames. In this paper, we propose an interpolation technique to distinguish foreground background to realize the error concealment of 3DV. First, we gate the frame difference between the front and rear frames. Limit comparison. The foreground and background are preliminarily divided, and further denoising is carried out by morphological filtering. For the background part, the interpolation is still carried out by the front and rear frames. For the foreground which is moving intensely, we use the adjacent viewpoint to fill the virtual composite image of the current viewpoint. The proposed method synthetically utilizes the correlation between the time domain and the viewpoint, and improves the quality of the reconstructed frame.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81

【相似文献】