Game Development Reference
In-Depth Information
ringing artifact. In addition, the motion compensated blocks generated by copying
interpolated pixel data from different locations of possibly different reference frames
may also incur artifacts. The in-loop filtering cannot only improve the visual quality
of the current frame, but also significantly improve the coding efficiency by providing
high quality reference for subsequent coding frames. Although it is a useful coding
tool, it also brings high complexity both in computation and hardware implemen-
tation. Therefore, until 1998, an in-loop filter (named as deblocking filter) was first
standardized in video coding, H.263v2 Annex J (H263
1998
). It was also extensively
debated during the development of the AVC/H.264 standard. Although it was finally
standardized in AVC/H.264 after a tremendous effort in speed optimization of the
filtering algorithm, the filter also accounts for about one-third of the computational
complexity of a decoder, which requires lots of conditional processing on the block
edge and sample levels.
Thanks to the improvement of computing capability, some more complex in-
loop filters can be integrated into video coding systems. In the development of
HEVC/H.265 and AVS2, two in-loop filters, Sample Adaptive Offset (SAO)
(Fu et al.
2012
) and Adaptive Loop Filter (ALF) (Tsai et al.
2013
), are extensively
discussed. The SAO reduces the compression artifacts by first classifying recon-
structed samples into different categories, obtaining an offset for each category, and
then adding the offset to each sample. Compared to SAO with only one offset for
each sample, the ALF processes one sample with neighboring samples by a multiple
taps filter, parameters of which are obtained by minimizing the distortion between
the distorted reconstruction frame and the original frame. Many ALF related tech-
niques are proposed during HEVC/H.265 development, e.g., Quadtree-based ALF
(Chen et al.
2011
) and LCU-based ALF (Tsai
2012
).
2.4 Quality Measurement
Video quality measurement is an important issue in video applications, and it also
plays an important role in the coding tools development. In general, video quality
assessment methods can be classified into subjective and objective quality assessment
two categories.
Subjective quality assessment can decide the final quality perceived by the human
through a subjective test. There are enormous subjective quality assessment methods.
In ITU-R BT.500-13 (
2012
), double-stimulus impairment scale (DSIS) method and
the double-stimulus continuous quality-scale (DSCQS) method as well as alternative
assessment methods such as single-stimulus (SS) methods, stimulus-comparison
methods, single stimulus continuous quality evaluation (SSCQE) and simultaneous
double stimulus for continuous evaluation (SDSCE) method are standardized for the
quality assessment of television pictures. However, the subjective test usually costs
many human and material resources, thus it cannot be used in real-time applications.
Objective assessment methods usually predict the visual quality by mathematical
models which can be quantitatively calculated. PSNR (peak signal noise ratio) is
