Digital Signal Processing Reference
In-Depth Information
6. Add the cost calculated in step 4 up to the total cost
J
tot
(
i
) respectively.
7. Go to the next sample and repeat step 4 and step 5 until all samples are processed.
8. Find out the minimum total cost in
J
tot
(
i
) , 0 ≤
i
≤
N
s
-1, and then the corresponding
trellis path is optimal in rate-distortion sense.
The PRDOTCQ algorithm has been designed for optimally quantizing the input sig-
nals by adaptively optimizing trellis path. Moreover it introduces JND threshold and
the perceptual weighted rate-distortion cost to acquire better subjective image quality.
This algorithm has no extra parameters and completely complies with the original
decoder. In addition, it can be applied to the TCQ-based image compression system.
4
Experimental Results
The PRDOTCQ algorithm has been verified on the platform of QTCQ system. Ex-
periments are performed on standard 512×512 greyscale, Peppers and baboon images
at several bit rates. In our experiment the processed window is defined as zone includ-
ing the first and second quantization cell. Since the first-order entropy is the good
estimate of the real coding bits with long input sequence, the average bit rate is esti-
mated by the first-order entropy [7].
The performance curves of PRDOTCQ, RDOTCQ and SPIHT algorithm are
shown in Fig.2-3. It can be seen that PRDOTCQ algorithm outperforms SPIHT algo-
rithm for all test sequences and all rates considered. A maximum gain up to 0.6 dB
can be achieved respectively with the same rate compared with SPIHT. Moreover, the
performance of RDOTCQ algorithm is better than the purposed algorithm. It is rea-
sonable since the purposed algorithm considers the human visual characteristic in the
rate distortion optimization.
Fig. 2.
Comparison of RD performance for SPIHT, RDOTCQ and PRDOTCQ for “peppers”
512
×
512 format
