基于多分支网络的深度图帧内编码单元快速划分算法

3维高效视频编码(3D-HEVC)标准是最新的3维(3D)视频编码标准,但由于其引入深度图编码技术导致编码复杂度大幅增加。其中,深度图帧内编码单元(CU)的四叉树划分占3D-HEVC编码复杂度的90%以上。对此,在3D-HEVC深度图帧内编码模式下,针对CU四叉树划分复杂度高的问题,该文提出一种基于深度学习的CU划分结构快速预测方案。首先,构建学习深度图CU划分结构信息的数据集;其次,搭建预测CU划分结构的多分支卷积神经网络(MB-CNN)模型,并利用构建的数据集训练MB-CNN模型;最后,将MB-CNN模型嵌入3D-HEVC的测试平台,通过直接预测深度图帧内编码模式下CU的划分结构来降低CU划分复杂度。与标准算法相比,编码复杂度平均降低了37.4%。实验结果表明,在不影响合成视点质量的前提下,该文所提算法有效地降低了3D-HEVC的编码复杂度。     

采用灰度共生矩阵进行深度预判的3D-HEVC深度图帧内快速编码算法

针对3D视频的3D-HEVC编码标准以多视点纹理视频和深度视频格式进行编码,其深度图编码仍延续纹理视频编码的模式和编码尺寸遍历选择,使得3D-HEVC的编码复杂度居高不下。本文针对深度图帧内预测编码,采用灰度共生矩阵对深度图中的CTU进行计算,统计并分析其矩阵中非零值个数与CTU分割深度的关系,根据非零值个数分布规律,设定阈值,使得帧内编码时可以预判编码模块的分割深度,从而选择性跳过部分不同深度CU的帧内预测过程。经过HTM16.0测试平台的检验,本算法在全帧内编码模式下,测试序列合成视点比特率仅增加0.08%的同时,平均节省了16.8%的编码时间,与其他同类较新算法在HTM16.0平台上的性能比较也有一定的优势。      

结合深度信息的多视点视频编码快速模式选择算法

多视点视频加深度(MVD,multi-view video plus depth)的编码格式包含多个纹理视频序列及其对应的深度图,深度图与对应的纹理视频具有相似的边缘信息。传统的编码模式不考虑两者的联系,单独编码导致复杂度高、编码时间过长。因此,合理利用深度图与纹理图的相似性进行编码,可以有效降低编码复杂度,同时应该确保编码质量不受影响。本文利用深度图的这一特点辅助纹理视频的编码,提出一种帧间快速模式选择算法,充分利用深度图与纹理图之间的相似性,建立一种联合复杂度模型,根据模型得到每个宏块的复杂度。对于复杂度低的宏块,在计算率失真代价之前跳过编码中一些不必要的模式,从而降低编码复杂度。实验表明,本文提出的快速模式选择算法,在保证率失真性能基本不变的情况下,能减少60.57%的编码复杂度,并最高能减少80.64%的编码时间。     

面向三维高效视频编码的深度图错误隐藏

基于多视点视频序列视点内、视点间存在的相关性,并结合视点间运动矢量共享技术,该文提出一种面向3维高效视频编码中深度序列传输丢包的错误隐藏算法。首先,根据3D高效视频编码(3D-HEVC)的分层B帧预测(HBP)结构和深度图纹理特征,将深度图丢失块分成运动块和静止块;然后,对于受损运动块,使用结合纹理结构的外边界匹配准则来选择相对最优的运动/视差矢量进行基于位移矢量补偿的错误掩盖,而对受损静止块采用参考帧直接拷贝进行快速错误隐藏;最后,使用参考帧拆分重组来获取新的运动/视差补偿块对修复质量较差的重建块进行质量提升。实验结果表明:相较于近年提出的对比算法,该文算法隐藏后的深度帧平均峰值信噪比(PSNR)能提升0.25～2.03 dB,结构相似度测量值(SSIM)能提升0.001～0.006,且修复区域的主观视觉质量与原始深度图更接近。     

一种HEVC帧内快速编码算法

高效视频编码(HEVC)采用编码单元(CU)四叉树的 分割结构,相比H.264/AVC显著地提升了编码效 率,但却使编码复杂度急剧增加。为此,本文提出一种帧内快速编码算法。首先,根据视 频图像纹理复 杂度,提前判断是否进行最大编码单元(LCU)分割。然后,根据空域相邻CU的深度预测当前C U的深度范围, 跳过不必要的计算;最后,根据预测模式被选为最优预测模式的统计特性,去掉可能性小的 帧内预测模式。本文算法在HM14.0的基础上实现。 仿真结果表明,本文算法在全I帧模式下与HM14.0相比,帧内编码时 间平均减少38%,码率(BR)只增加1.41%,峰值信噪比(PSNR)只降低0.29dB,在保证编码性能和视频质量几乎不变的 情况下,本文算法降低了编码的计算复杂度。     

基于相似度的HEVC帧内低复杂度算法

针对高清视频编码(HEVC)标准帧内编码计算 复杂度过高的问题,提出了基于最大编码单元(LCU)相似度的 深度范围及率 失真代价阈值预测低复杂度算法。首先分析LCU相似性并将其量化,然后利用该量化变量概 率密度分布规律, 采用贝叶斯分类方法估计当前编码LCU深度范围;同时考虑到编码单元(CU)之 间相关性,自 适应设定率失真代价阈值,用以提前结束模式选择及继续分割过程:通过LCU编码深度范围 预测和率失真代 价阈值实现快速CU划分方案。实验结果表明,在性能几乎不变的情况下,本文算法较标准 编码算法编码 时间平均减少了39.55%,平均BDBR增加在1%以内,BD-PSNR仅降低了 0.03dB,在保持良好视频质量下有效降低了编码复杂度。     

3D-HEVC合并模式快速判决方法研究

基于高效视频编码的3D视频编码（3D-HEVC）是目前正在研究的新一代3D视频编码标准。为降低3D-HEVC中模式选择的计算复杂度,根据非独立视点纹理图中合并模式采用率高的特点,该文提出了一种3D-HEVC合并模式快速判决方法。在B帧中,分析了当前编码单元（CU）与视点方向参考帧中参考块间编码模式的相关性;在P帧中,分析了位于相邻划分深度的CU间编码模式的相关性。根据分析的视点间和划分深度间的相关性设计快速判决条件,预判采用合并/合并-跳过模式编码的CU,判别出的CU在模式选择过程中只检查相关的候选预测模式,从而降低计算复杂度。实验结果表明,与3D-HEVC原始算法相比,该文算法能够在率失真性能损失很小的前提下,平均节省11.2%的总编码时间和25.4%的非独立视点纹理图的编码时间。      

一种HEVC的快速帧间编码新方法

为了降低新一代高效视频编码(HEVC)的计算复杂度 ,融合基于人眼恰可察觉失真(JND)的快速帧间 模式选择和自适应编码单元深度范围(ACUDR)两种算法,提出了一种快速帧间 编码新方法。首先,通过建立时空域JND和最佳预测模式的统计关系,利用JND 阈 值指导帧间预测模式的快速选择,以减少遍历的模式个数从而降低复杂度;然后,利用编码 的前一帧定义与当前帧 的平均深度差(ADD),通过ADD自适应加权时空域相邻最大CU(LCU) 的平均深度值预测当前LCU的深度范围(DR)类型,减少当前LCU遍 历的深度个数从而降低算法复杂度; 最后,融合这两种算法以进一步降低了视频编码的计算复杂度。结果表明,所提出方法 在低延时和随机访问两种 编码结构下,降低了40%以上的计算复杂度,但编码性能降低很小,在同等码率下的峰值信 噪比(BDPSNR)仅分别降低0.058dB和0.037dB,同等客观质量下 的码率(BDBR)仅分别增加1.71%和1.06%。     

纹理-深度图共同优化的3D-HEVC帧内快速算法

随着立体及3D视频需求的日益增多,针对3D视频编码方法的研究受到越来越多的关注。3D-HEVC编码标准对采用纹理和深度图格式融合的3D视频进行编码,由于加入了深度图编码,因此新增了深度图编码模式、组件间预测和分段直流编码等技术,使其编码复杂度急剧升高。为了减少3D-HEVC的编码时间,本文提出了针对纹理图和深度图的编码单元（Coding Unit,CU）尺寸提前决策快速算法。利用梯度矩阵和作为当前CU和子CU复杂度的判断依据,将CU分为三类：不划分CU(Non-Split Coding Unit,NSCU)、直接划分CU(Split Coding Unit,SCU)以及普通CU。对NSCU,跳过小尺寸的帧内预测过程;对SCU,直接跳过当前CU的帧内预测过程;对普通CU,执行原平台操作。实验结果表明,与原始平台相比,本文算法在合成视点质量基本不变的情况下,平均减少40.92%的编码时间;与最新的联合纹理-深度图优化的3D-HEVC快速算法相比,可以在质量相当的情况下减少更多的编码时间。     

基于多视点视频特征的3D-HEVC深度视频信息隐藏算法

为了更安全地传输秘密信息和保护3D视频,提 出基于多视点视频特征的3D-HEVC深度视频信息隐 藏算法。首先结合彩色视频的纹理特征,考虑深度视频不同区域对绘制视点的质量和编码效 率的影响,对 深度视频进行区域分割。然后针对不同的区域,采用不同的方式调制最大编码单元(LCU) 的QP值嵌入 秘密信息。最后,用修改后的QP值进行编码压缩,传输视频信息。实验仿真结果表明,相 比于原始HTM13.0, 本算法编码重建深度视频绘制视点质量的PSNR平均下降0.0015dB, 码率平均增加0.035%。本文算法 能较好地保证绘制视点的质量,对视频流的码率影响较小,且能实现秘密信息的盲提取。     

Multi-class support vector machine-based fast algorithm for 3D-HEVC depth video intra coding

The recursive splitting process of largest coding unit (LCU) and the mode search process of coding unit imposed enormous computational complexity on encoder. A multi-class support vector machine-based (MSVM) fast coding unit (CU) size decision algorithm for 3D-HEVC depth video intra-coding was proposed. The algorithm included two steps: off-line training and fast CU size and mode decision. In the process of off-line training, a MSVM model was constructed, where the texture complexity of current LCU, the optimal partition depth of its spatial neighboring LCU and co-located LCU in texture video were treated as feature vectors, and the optimal partition depth of LCU was utilized as corresponding class label. In the process of fast CU size and mode decision, features of LCU were extracted before cod-ing a LCU, then, a MSVM model was used to predict the class label. Finally, the class label that represents the largest parti-tion depth of the current LCU was employed to terminate the CU recursive splitting process and CU mode search process. Experimental results show that the proposed algorithm saves the encoding time of 3D-HEVC by 35.91% on average, and the encoding time of depth video by 40.04% on average, with negligible rendered virtual view image degradation.     

An image feature-based method to efficiently determine inter-coding depth in HEVC

The quad-tree based picture partition scheme in High Efficiency Video Coding (HEVC) results in a more substantial increase in computational complexity than those incurred by its predecessor video coding standards because of the need in this scheme to determine the best coding unit (CU) partitions. In this paper, we propose a method to effectively reduce the computational complexity of inter-prediction coding in the HEVC standard. The relative displacement of the largest coding unit (LCU) at the corresponding position between adjacent frames is tested through optical flow (motion estimation). The texture intensity of the LCU at the given time is tested if the condition that determines the coding depth in advance cannot be satisfied. The depth of the coding unit (CU) can be determined in advance beyond the xCompressCU function by using our proposed method, which does not require the calculation of the rate-distortion (RD) cost for each level of depth, and thus reduces the circular traversal times of the xCompressCU function. Experimental results proved that our proposed method is effective, as it reduced the computational complexity of an encoder by 53.2% on average, and had a slight influence on coding performance.     

一种低复杂度的HEVC帧内快速编码算法

为了降低高效视频编码(HEVC)帧内编码复杂度,提出一种HEVC帧内快速编码算法。根据视频图像的纹理复杂性,提前跳过或者中止部分尺寸的编码单元(CU)的划分,减少CU深度遍历区间;同时,根据粗选过程后预测模式和代价值的统计特性采用阈值法或者梯度模式直方图法进一步筛选掉粗选后可能性较小的预测模式,从而减少最后进行率失真(RD)代价计算的帧内预测模式数量,进一步降低编码复杂度。实验结果表明,本文算法与HEVC原始平台相比,在全I帧编码模式下编码时间平均减少42.20%,码率(BR)上升约1.75%,峰值信噪比(PSNR)降低了0.108dB,有利于实时应用。     

Fast prediction unit selection method for HEVC intra prediction based on salient regions

In order to reduce the computational complexity of the high efficiency video coding(HEVC) standard, a new algorithm for HEVC intra prediction, namely, fast prediction unit(PU) size selection method for HEVC based on salient regions is proposed in this paper. We first build a saliency map for each largest coding unit(LCU) to reduce its texture complexity. Secondly, the optimal PU size is determined via a scheme that implements an information entropy comparison among sub-blocks of saliency maps. Finally, we apply the partitioning result of saliency map on the original LCUs, obtaining the optimal partitioning result. Our algorithm can determine the PU size in advance to the angular prediction in intra coding, reducing computational complexity of HEVC. The experimental results show that our algorithm achieves a 37.9% reduction in encoding time, while producing a negligible loss in Bjontegaard delta bit rate(BDBR) of 0.62%.     

Self-learning residual model for fast intra CU size decision in 3D-HEVC

As an extension of the High Efficiency Video Coding (HEVC) standard, 3D-HEVC requires to encode multiple texture views and depth maps, which inherits the same quad-tree coding structure as HEVC. Due to the distinct properties of texture views and depth maps, existing fast intra prediction approaches were presented for the coding of texture views and depth maps, respectively. To further reduce the coding complexity of 3D-HEVC, a self-learning residual model-based fast coding unit (CU) size decision approach is proposed for the intra coding of both texture views and depth maps. Residual signal, which is defined as the difference between the original luminance pixel and the optimal prediction luminance pixel, is firstly extracted from each CU. Since residue signal is strongly correlated with the optimal CU partition, it is used as the feature of each CU. Then, a self-learning residual model is established by intra feature learning, which iteratively learns the features of the previously encoded coding tree unit (CTU) generated by itself. Finally, a binary classifier is developed with the self-learning residual model to early terminate CU size decision of both texture views and depth maps. Experimental results show the proposed fast intra CU size decision approach achieves 33.3% and 49.3% encoding time reduction on average for texture views and depth maps with negligible loss of overall video quality, respectively.     

一种基于视频内容特征的LCU层码率控制算法

针对高效率视频编码LCU层码率控制方案对目标 比特分配不合理,图像复杂度综合 因子考虑不全导致率失真性能损失较大的问题,提出了一种基于视频内容特征的LCU层码 率控制算法。通过Scharr算子和Hadamard变换分别检测编码区域的纹理复杂度和能量分 布;对纹理复杂度因子和能量占比因子进行联合加权,构建了一种新的权重计算公式,用于 LCU层目标比特的优化分配,并进行模型参数的调整。实验结果表明,本文算法在低延迟B 配置和P配置下,与动态调整码率控制算法相比,比特率相对误差平均下降了0.011%和0.010%,率失真性能平均提高了2.5%。