Digital Signal Processing Reference
In-Depth Information
A Perceptual Weighted Trellis-Coded
Quantization Algorithm
Wei Jiang, Junjie Yang, and Jason Gao
School of Computer and Information Engineering,
Shanghai University of Electric Power, Shanghai 200090, China
zzm7406@sina.com
Abstract.
The rate-distortion optimization algorithm (RDOTCQ) was proposed
to improve the coding performance efficiently in our former work. However,
human visual characteristics are not considered. In this paper, a perceptual
weighted rate-distortion optimization TCQ (PRDOTCQ) algorithm is proposed.
By introducing the vision model that takes into account various masking effects
of human visual perception and a perceptual distortion metric, the proposed
algorithm obtains better subjective quality. The experiment shows that
PRDOTCQ algorithm provides better performance compared to SPIHT with the
maximum gain 0.6dB. Though it behaves a little PSNR loss compared with
RDOTCQ, it results in better subjective quality with the same rate.
Keywords:
Rate-distortion, JND, TCQ.
1
Introduction
Trellis-Coded Quantization (TCQ) stems from trellis-coded modulation. By using the
expanded signal set and set partitioning ideas it can be thought of being a type of
vector quantization [1]. Because of its excellent MSE performance and moderate
complexity, TCQ-based schemes have been widely applied to image compression [2]-
[4]. The entropy-constrained TCQ (ECTCQ) is introduced in [5]-[6]. Though these
ECTCQ systems achieve good performance, it is difficult to be implemented because
separate TCQ codebooks must be precomputed for each encoding rate. Based on rate-
distortion criterion, a rate-distortion optimized Trellis-Coded Quantization
(RDOTCQ) algorithm is presented [7]. By introducing the rate-distortion cost into the
trellis path selection it can improves the performance with a little computational com-
plexity increase. These algorithms aim at improvement of coding efficiency in infor-
mation theory. Since most images will be perceived by human consumers, encoding
and transmission of the unaware data is a waste of bitrates. Therefore, there is a large
room for coding efficiency improvement by reducing the perceptual data redundancy
according to human visual theory.
In this paper, a perceptual weighted rate-distortion optimization trellis-coded quan-
tization algorithm (PRDOTCQ) is proposed. The human visual theory is applied to
get rid of the unaware data. Moreover, it optimizes the decisions on how to map input
