Game Development Reference
In-Depth Information
In the development of video coding standards, many perceptual rate
distortion optimization techniques have been proposed. In 1993, perceptual model
(Webster et al.
1993
) has been applied to MPEG-1 video coding standard to improve
the coding performance. Recently, due to the best tradeoff among accuracy, simplic-
ity, and efficiency of SSIM index, perceptual rate distortion optimization techniques
based on the SSIM index have been proposed. In Mai et al. (
2006
), SSIM index was
applied in themotion estimation process to select the best motion vectors. InHuang et
al. (
2010
), Wang et al. (
2012
), SSIM index was applied in the mode decision process
to obtain the best coding mode. In order to achieve the optimal RD performance, it is
very important to carefully choose
ʻ
value. As SSIM is highly adaptive to the input
content, constant
ʻ
is not an optimal solution in perceptual RDO process. To derive
the adaptive Lagrangian multiplier, in Huang et al. (
2010
), the key frame is coded
in two passes, so that the difference between rate and distortion can be employed in
estimating
ʻ
. In Wang et al. (
2012
), reduced reference distortion and rate models are
established to estimate the Lagrangian multiplier.
By locally adjusting the Lagrangian multiplier, MB level rate control can be
achieved, as smaller
ʻ
corresponds to the coding mode with larger coding bits, and
vice versa. In Sun et al. (
2008
), the researchers proposed a visual sensitivity model
with an attentionmodel, a positionmodel and a texture structuremodel. Then adaptive
Lagrangian multiplier adjustment scheme is proposed to achieve best rate allocation
in MB level. In Pan et al. (
2005
), adaptive Lagrangian multiplier selection scheme
is proposed in scalable video coding.
12.2.2.4 Preprocessing and Postprocessing
Though the preprocessing, in-loop filter, and postprocessingmodules are not included
in the actual compression process, they play critical roles in improving the perceptual
quality of the reconstructed video.
Attention-based preprocessing scheme for video coding was proposed in Itti
(
2004
). Based on the biologically motivated attention characteristics of HVS, each
input frame can be blurred with a dynamic foveation filter. Then with MPEG-1 and
MPEG-4 encoding, substantial coding gain can be achieved.
In video coding, different kinds of artifacts are introduced, such as blocking,
blurring, ringing and temporal fluctuations. Postprocessing techniques can be applied
to reduce these artifacts. The main issue is that when reducing on kind of artifacts,
another kind of artifacts is simultaneously introduced. Therefore, it is important to
balance these coding artifacts from the perceptual quality point of view. Recently,
postprocessing techniques based on perceptual cues have been proposed (Chetouani
et al.
2009
; Liu et al.
2010
), which can efficiently reconstruct the distorted image
with high perceptual quality. In addition, learning-based image restoration scheme
for compressed images was proposed inMa et al. (
2008
), inwhich the high-frequency
components with the priors are learned from a training set of natural images.
