Overview of the High Efficiency Video Coding Standard——Inter-picture prediction(一)

来源:互联网 发布:中原工学院软件类专业 编辑:程序博客网 时间:2024/05/22 13:18

G. Intra-picture prediction 

For intra prediction, previously decoded boundary samples from adjacent PUs must be used.Directional prediction with 33 different directional orientations is defined for (square) PU sizes from 4×4 up to 32×32. The possible prediction directions are shown in Fig. 6; alternatively, planar prediction (assuming an amplitude surface with a horizontal and vertical slope derived from the boundaries) and DC prediction (a flat surface with a value matching the mean value of boundary) can also be used. For chroma, the horizontal, vertical, planar, and DC prediction modes can be explicitly signaled, or the chroma prediction mode can be indicated to be the same as the luma prediction mode (and, as a special case to avoid redundant signaling, when one of the first four choices is indicated and is the same as the luma prediction mode, the Intra_Angular[ 34 ] mode is applied instead). 

Each CB can be coded by one of several coding types, depending on the slice type. Similar to H.264/MPEG-4 AVC, intra coding is supported in all slice types. HEVC supports various intra coding methods referred to as Intra_Angular, Intra_Planar and Intra_DC. The following subsections present a brief further explanation of these and several techniques to be applied in common.



帧内预测是以PU尺寸进行操作的;并且在空域上相邻的前面已解码的边界像素将被用作预测参考信号;对从4x4到32x32的PU定义了33个不同的预测方向;所有可能的预测方向如图6中所示;而且Planar预测和DC预测同样可以使用;对于色度分量,水平,垂直,planar,DC预测模式可以显示使用;也可以直接沿用相应的亮度分量的预测模式;

根据片类型的不同,每个CB可以依据不同的编码类型来进行编码。和H.264/MPEG-4 AVC一样,帧内预测编码支持所有的片类型;HEVC支持多种帧内预测编码方法,包括方向,planar, DC;下面将对通常使用的技术做进一步的解析;

补充:目前HM模型中共包含了35种预测模式。但由于受到编码复杂度限制,编码模型对4×464×64尺寸的PU所能使用的预测模式进行了限制。色度分量的帧内预测采用了5种预测模式,分别为水平、垂直、DC预测、亮度模式以及对角模式。




1) PB Partitioning: 

An intra-coded CB of sizeM×M may have one of two types of PB partitions referred to as PART_2N×2N and PART_N×N, the first of which indicates that the CB is not split and the second indicates thatthe CB is split into four equal-sized PBs. (Conceptually, in thisnotation, N = M/2.) However, it is possible to represent thesame regions that would be specified by four PBs by usingfour smaller CBs when the size of the current CB is largerthan the minimum CU size. Thus, the HEVC design onlyallows the partitioning type PART_N×N to be used when thecurrent CB size is equal to the minimum CU size. This meansthat the PB size is always equal to the CB size when theCB is coded using an intrapicture prediction mode and theCB size is not equal to the minimum CU size.

一个尺寸为MxM的帧内编码CB,有两种PB分割的形式:PART_2Nx2N和PART_NxN。第一种,PART_2Nx2N,表示这个CB不能再被划分;第二种类型,PART_NxN,  表示这个CB可以再被划分成四个相同尺寸(N = M/2)的PB,如上图补充图所示;然而,当当前CB尺寸大于最小的CU尺寸时,四个PB会使用四个更小的CB来表示相同区域;因此,当当前CB的尺寸等于最小的CU尺寸时,HEVC只允许PART_NxN划分类型这意味着,当CB使用帧内预测编码且CB尺寸不等于最小CU尺寸时,PB尺寸问题等于CB尺寸;

2)Intra_Angular prediction: 

Spatial-domain intra prediction has previously been successfully used in H.264/MPEG-4 AVC. The intra prediction of HEVC is also based on spatial-domain intra prediction, but is extended significantly – mainly due to the increased size of the PB and an increased number of selectable prediction directions. Compared with 8 directional intra predictions of H.264/MPEG-4 AVC, HEVC supports a total of 33 directional intra predictions denoted as Intra_Angul/ar[ k ] where k is a mode number from 2 to 34. The angles are intentionally designed to provide denser coverage for near-horizontal andnear-vertical angles and coarser coverage for near-diagonal angles to reflect the observed statistical prevalence of the angles and the effectiveness of the signal prediction processing. 

空域帧内预测在 H.264/MPEG-4 AVC上有很多成功的应用。HEVC在帧间的预测也是基于这种空域的帧间预测,但它研究起来更需注意,主要是因为PU尺寸的增长以及可选预测方向的增长。相对于H.264/MPEG-4 AVC中的8个帧内预测方向,HEVC支持33个预测方向,总称为Intra_Angular[k], 其中k为2到34;通过对这些角度的选取提供了一个覆盖相邻的水平和垂直边界密集的范围以及粗略的对角线位置来反映角度在统计学上的普遍性和信号预测处理的时效性;
When using an Intra_Angular mode, each PB is predicted directionally from spatially neighboring samples which are reconstructed (but not yet filtered by the in-loop filters) before being used for this prediction. For a PB of size N×N, a total of 4N+1 spatially neighboring samples may be used for the prediction, as shown in Fig. 6. When available from preceding decoding operations, samples from lower left PBs can be used for prediction in HEVC in addition to samples from PBs at the left, above, and above-right of the current PB. The prediction process of the Intra_Angular modes can involve extrapolating samples from the projected reference sample location according to a given directionality. To remove sample-by-sample switching between the reference row and column buffers, all extrapolations in a PB refer to a single reference row or column depending on the mode number. For Intra_Angular[ k ] with k in the range of 2 to 17, the samples located at the left column are used for the extrapolation, and the samples located at the top row are used when k is in the range of 18 to 34. 当使用角度预测模式时,每个PU都使用空域上相邻的已重建的像素来预测;对于一个N×N大小的PU,总共有4N+1个空域相邻的像素被用于预测如上图6所示,当先前PU解码操作有效时左下角相邻PU,左相邻PU,上和右上相邻TB的相邻边界像素均可用作预测;ntra_Angular模式的预测处理可以从给定的方向的参考像素位置推算得出,为了逐个像素点在参考行和列的交换,所有在PB的推算都是基于模式数的单行或列。对于k从2~7,位于上行的像素被投影成位于左列的额外像素;k为18-34时,位于左列的像素被投影成位于上行的像素;(这东西不太好理解)

To improve the intra prediction accuracy, the projected reference sample location is computed with 1/32 sample accuracy. Bi-linear interpolation is used to obtain the value of the projected reference sample using two closest reference samples located at integer positions. The prediction process of the Intra_Angular modes is consistent across all block sizes and prediction directions, whereas H.264/MPEG-4 AVC uses different methods for its supported block sizes of 4×4, 8×8 and 16×16. This design consistency is especially desirable since HEVC supports a greater variety of PB sizes and a significantly increased number of prediction directions compared to H.264/MPEG-4 AVC. 

为了提高帧内预测的精度,被投影的参考像素精度位置使用1/32像素精度;为了获得被投影的参考像素,需要对位于整数位置间的两个最近的参考像素使用线性插值;Intra_Angular模式的预测处理对于所有块尺寸和预测方向都具有一致性;而H.264/MPEG-4 AVC对不同的块尺寸(4x4,8x8,16x16)使用不同的方式;相比较H.264/MPEG-4 AVC. ,HEVC的这个设计很有意义,因为很适合大的PB尺寸以及增加的多种预测方向;

3)Intra_Planar and Intra_DC prediction :

In addition to Intra_Angular prediction which targets regions having strong directional edges, HEVC supports two alternative prediction methods, Intra_Planar, and Intra_DC, for which similar modes were specified in H.264/MPEG-4 AVC. While Intra_DC prediction uses an average value of reference samples for the prediction, average values of two linear predictions using four corner reference samples are used in Intra_Planar prediction to prevent discontinuities along the block boundaries. The Intra_Planar prediction mode is supported at all block sizes in HEVC while H.264/MPEG-4 AVC only supports plane prediction when the luma prediction block size is 16×16, and its plane prediction operates somewhat differently from the planar prediction in HEVC. 

除了角度预测这种目标区域有较强的方向边界,HEVC还有其他两种预测方式。Intra_Planar和Intra_DC两种模式在H.264/MPEG-4 AVC中都有类似的说明。Intra_DC预测模式使用参考像素的均值进行预测,而Intra_Planar则是用四个角的参考像素得到的两个线性预测的均值,这样能有效地防止块边界的不连续情况。Intra_Planar模式在HEVC中支持所有尺寸大小的块,而在 H.264/MPEG-4 AVC中只有plane预测模式且中能于大小为16x16的亮度块,且它们的方法也不一样;
补充:Planar 模式在平面预测中,当前块右角的像素值保存在比特流中,其最右边和底部的像素由线性插值得到,而其余的像素通过边界的像素双线性插值得到。

4)Reference sample smoothing :

In HEVC, the reference samples used for the intra prediction are sometimes filtered by a 3-tap [1 2 1]/4 smoothing filter, in a manner similar to what was used for 8×8 intra prediction in H.264/MPEG-4 AVC. However, HEVC applies this smoothing operation more adaptively according to the directionality and the block size. As in H.264/MPEG-4 AVC, the smoothing filter is not applied for 4×4 blocks. For 8×8 blocks, only the diagonal directions, Intra_Angular[k] with k = 2, 18, or 34, use the reference sample smoothing. For 16×16 blocks, the reference samples are filtered for most directions except the near-horizontal and near-vertical directions, k in the range of 9 to 11 and 25 to 27. For 32×32 blocks, all directions except the exactly-horizontal (k = 10) and exactly-vertical (k = 26) directions use the smoothing filter. The Intra_Planar mode also uses the smoothing filter when the block size is equal or greater than 8×8, and the smoothing is not used (or useful) for the Intra_DC case. The Intra_Planar mode also uses the smoothing filter when the block size is equal or greater than 8×8, and the smoothing is not used (or useful) for the Intra_DC case. 

在HEVC中,用于帧内预测的参考像素点有时候用三阶([1 2 1]/4)平滑滤波器来平滑,这和H.264/MPEG-4 AVC中的8×8帧内预测模式一样。然而HEVC采用的这种平滑方式更适应于方向性和块的大小。H.264/MPEG-4 AVC中,平滑滤波不适用于 4×4块。对于 8×8块,只有在对角线方向,Intra_Angular[k] with k = 2, 18, or 34,用参考像素平滑。对于16×16块,大部分的方向除了接近水平和垂直方向( k in the range of 9 to 11 and 25 to 27)都使用参考像素平滑;对于 32×32块,除了恰好水平K=10和恰好垂直K=26其他方向都使用平滑滤波。The Intra_Planar块大小大于或等于8×8时,也使用平滑滤波,对于Intra_DC的模式,平滑不用。

5)Boundary value smoothing :

To remove discontinuities along block boundaries, in three modes, Intra_DC (mode 1) and Intra_Angular[k] with k = 10 or 26 (exactly-horizontal or exactly-vertical), the boundary samples inside the prediction block are replaced by filtered values. For Intra_DC mode, both the first row and column of samples in the PB are replaced by the output of a 2-tap [3 1]/4 filter fed by their original value and the adjacent reference sample. In horizontal (Intra_Angular[ 10 ]) prediction, the boundary samples of the first column of the PB are modified such that half of the difference between their neighbored reference sample and the top-left reference sample is added. This makes the prediction signal more smooth when large variations in the vertical direction are present. In vertical (Intra_Angular[ 26 ]) prediction, the same is applied to the first low of samples. 
为了消除块边界处的不连续,在三种模式下(DC模式,垂直以及水平方向模式)边界的像素值被滤波值取代。 Intra_DC模式,PU内的第一行和第一列的像素将被以原像素和相邻像素作为二阶([3 1]/4)滤波器输入的输出值代替;在水平方向模式上,PB第一列的边界像素将修改为他的邻近参考像素的差值和左上参考像素的和的一半;这使得预测信号在垂直方向变化明显时信号更加平滑;Intra_Angular[ 26 ]和水平的同样理解;    

原创粉丝点击