关于3D-HEVC中的SDC编码方式

在阅读3D-HEVC的相关文献时，经常会读到SDC这种编码方式，有的文献将其写成“Simplified Depth Coding”，也有的文献将其写成“Segment-wise DC Coding”。

这部分一直令我很晕，因此趁今天有时间，就对关于SDC的提案进行了查找和研究并在此进行一个简单的总结。

关于SDC的发展过程是这样的：

SDC是在JCT3V的第二次会议中被首次提出的，当时的名称是“Simplified Depth Coding”，具体见提案（JCT3V-B0036）。

在SDC被提出之后，经过了3次会议的不断发展和改进（第3次到第5次），直到第6次会议时，该算法基本定型，并且将其改名为“Segment-wise DC Coding”。

因此，在第6次会议之前的论文中使用的都是旧的名称，而在第6次会议之后的文献中就使用的是新的名称。这两种叫法指的都是同一个算法。

下面是SDC算法的原理，我这里写的是第11次会议提案中对SDC的描述（JCT3V-K1003），后面会议的提案还没有公布，不过对于SDC应该不会有什么变化了。

这部分原文很好理解，读起来也很简单，我就不翻译了，直接把原文贴上。

Segment-wise DC coding

The Segment-wise DC Coding (SDC) approach provides an alternative residual coding method. With SDC, the residual data (one or two constant residual values within one PU) is coded without transform and quantization processes. Whether SDC is used is signalled in the coding unit parameters structure as described in section 2.3.5 at PU level. The partition size of CU containing a SDC coded PU is always 2Nx2N. SDC can be applied to all depth intra prediction modes including HEVC intra prediction modes and depth modelling modes. For HEVC intra prediction modes, the entire PU is considered as one segment, while for DMM modes, there are two segments.

Before coding, the residual values are mapped to values, which are present in the original, uncompressed depth map by using a Depth Lookup Table (DLT). Consequently, residual values can be coded by signalling only the index into this lookup table, which reduces the bit depth of residual magnitudes. This mapping table is transmitted to the decoder for the inverse lookup from index to valid depth value.

The advantage of using this lookup table is the reduced bit depth of the residual index for sequences with reduced depth value range (e.g. all estimated depth maps where not all depth values are present).

At encoder side SDC process utilizes the mean of the original depth value (d_orig) and the predicting depth value (d_pred). As illustrated in the example shown in Figure 28, for SDC (HEVC intra prediction modes), d_pred is calculated as the average of the left-top, right-top, left-bottom, and right-bottom samples in a predicted block. For SDC (DMM Mode 1 and DMM Mode 4), d_pred of each segment is the same as the CPV predictor generated from the section 2.3.3.3e.

↑这个图是HEVC intra prediction modes

↑这个图是DMM Mode 1

后面还有几个公式我就以截图的方式贴上了，敲公式太费劲了。

以上就是关于SDC的内容，当然，关于DLT的建立过程提案中也有详细的讲解，感兴趣朋友的可以自己下载提案看一看。