Improved Sample Adaptive Offset for HEVC 阅读笔记

本文主要是阅读HEVC-SAO相关的paper，了解算法原理。

术语+链接

SAO sample adaptive offset (SAO)

SAO是什么作用是什么？

> The deblocking filter (DBF) is applied to the bound-
aries of reconstructed block to reduce the blocking artifacts.
In addition to DBF, a new technique called sample adaptive
offset (SAO) filter is applied adaptively to the reconstructed
samples after DBF

>SAO is helpful to reduce the ringing
arifacts, which mainly come from the quantization errors of
transform coefficients.

ringing arifacts 响铃效果如下图

https://en.wikipedia.org/wiki/Ringing_artifacts

具体怎么减少该影响？

 >In HEVC, SAO is an in-loop filter and located after de-
blocking filter as depicted in Fig.1. The idea of SAO is to
compensate reconstructed samples by adding an offset to each
pixel, so that the distortion between reconstructed picture and
original one can be reduced

通过对每个像素添加偏移量来补偿重建样本，从而减少重建图像与原始图像之间的失真。

>The key problem of this method
is how to classify the reconstructed samples and how to select
the offsets for each category.

 In current SAO, two different
method: band offset (BO) and edge offset (EO) are selected.

EO

1-D Edge分为四类

四种类别的有5中采样策略

统计分析发现，1，2两种策略offset为正数，3，4两种为负数。这一发现有效减少编码符号的bit数量。

 

BO

总结

改论文核心观点是：使用多种EO类型组合起来能达到最好的效果，当知道编码序列的特征时，可以按需选择相应的组合方式。论文中0&1组合适用于大多数的测试场景

参考文献

1-D Edge 一维边缘检测

https://cs.gmu.edu/~kosecka/cs482/cs482-edges.pdfhttps://cs.gmu.edu/~kosecka/cs482/cs482-edges.pdf