Game Development Reference
In-Depth Information
average image and the difference images between central and non-central images
are compressed individually. Yue et al. ( 2012 ) proposed to only compress image
thumbnail and image local feature descriptor, and reconstruct high quality image
with similar image patches retrieved from cloud. Extending it to video compression,
Wang et al. ( 2014 ) imitated the multi-view coding method to jointly compress several
near-duplicate videos by referring the video itself or other coded similar videos.
2.3.2 Transform Coding
Besides the prediction coding, transform coding can reduce the correlation existing
in the prediction residual signal through transforming the signal from spatial domain
to the frequency domain by orthogonal transform. In 1965, Enomoto and Shibata
( 1965 ) first proposed a practical video compression method with one dimensional
Hadamard transform. In 1968, Andrews and Pratt ( 1968 ) extended the Enomoto's
method to two dimensional transform with block-based Fourier transform. How-
ever, the Fourier transform has a relative high computational complexity, which
is difficult to apply in practical video coding system. In 1973, Chen ( 1973 )pro-
posed Slant transform and the corresponding fast computational algorithm, which
has higher energy compaction property than Fourier transform. In 1974, Ahmed
et al. ( 1974 ) proposed the Discrete Cosine Transform (DCT), which has much lower
computational complexity, and is more effective in decorrelation and energy con-
centration. In 1979, Netravali and Stuller ( 1979 ) proposed motion compensation
transform framework, which is well known as hybrid prediction/transform coder
nowadays and called first generation coding methods usually, and has been widely
used in video coding standards since H.261.
In AVC/H.264, integer transform is developed to do DCT-like transformwith sim-
ple integer arithmetic (Bjontegaard 1997 ;Malvaretal. 2003 ), which can also avoid
the inverse-transform mismatches. Considering the efficiency of different transform
block size, the large transforms can provide a better energy compaction and a better
preservation of detail than small transforms but larger transforms introduce more
ringing artifacts caused by quantization than small transforms. The adaptive block-
size transform (ABT) (Wien 2003 ) is proposed in AVC/H.264 to improve the coding
efficiency, which applies the same transform size as the prediction block size instead
of only 4
4 transform. In the development of HEVC/H.265, more adaptive trans-
form schemes were studied, e.g., mode-dependent directional transform (MDDT)
(Ye and Karczewicz 2008 ), rate distortion optimized transform (RDOT) (Zhao et
al. 2012 ). Especially, according to the statistical characteristics of intra prediction
residual, Discrete Sine Transform (DST) was proposed for more efficient coding
(Saxena and Fernandes 2011 ).
×
 
Search WWH ::




Custom Search