Game Development Reference
In-Depth Information
In Wang et al. (
2012
), the structural information is accounted by two descriptors
extracted from the 3-D structure tensors, which are the largest eigenvalue as well as
its corresponding eigenvector. Noticing that the frame rate also has great impact on
the video quality, in Ma et al. (
2012
), the final quality is modeled as a function of
frame rate and quantization parameter.
12.2.2 Perceptual Video Coding
Based on these quality assessment models, a lot of effort has been devoted to devel-
oping perceptual video coding algorithms, which can efficiently improve the video
compression performance from the perceptual distortion-rate point of view. In the
hybrid video coding framework, the perceptual cues can be applied to many compo-
nents, such as prediction, quantization, rate distortion optimization, pre- and post-
processing, etc. It is noted that most of these techniques are encoder-optimization
tools, which satisfies the condition of standard (decoder) compatibility.
Some pioneer works on perceptual video coding are trying to break through the
traditional hybrid coding framework to improve the coding efficiency. The most rep-
resentative work is texture analysis and synthesis-based video compression (Ndjiki-
Nya et al.
2009
; Zhu et al.
2008
). This technique exploits the texture masking effect
in spatial and temporal domain to generate the texture region at the decoder. There-
fore, the texture region at the encoder can be automatically skipped. Based on this
idea, inpainting-based image compression schemes were also proposed in Liu et al.
(
2007
,
2008
). At the decoder, uncoded pixels can be generated from the inpainting
technique to improve the coding efficiency.
Herewe give a brief overviewof perceptual coding fromJND-based residual filter-
ing, perceptual quantization, perceptual distortion-rate optimization ,and perceptual
filtering in pre- or postprocessing.
12.2.2.1 JND-Based Residual Filtering
Near-threshold models can be effectively applied to the prediction loop, in which
the residuals are filtered with the precomputed JND. The JND-based video cod-
ing scheme is illustrated in Fig.
12.2
. Specifically, before coding a frame, the JND
Fig. 12.2
JND-based video coding scheme
