This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Medical Image Compression by Discrete Cosine Transform Spectral
Similarity Strategy
Source: IEEE Transactions on Information Technology in Biomedicine, Vol. 5, No. 3, Sept. 2001
Authors: Yung-Gi Wu and Shen-Chuan Tai
Speaker: Hsien-Chu Wu
Date: 02/21/2002
2
Outline
• Introduction
• Proposed strategy
• Simulation results
• Conclusions
3
IntroductionGoal: Develop a strategy to raise the compression ration by exploiting spectra similarity while preserving good decoded quality.
Observation:
4
Proposed Strategy
A. Spectrum Reorganization
Need a buffer to store all individual 88 DCT transformed data sequentially
/.index spectrum /1)-(N/b)(j,0 ,,
/,indexfrequency /1)-(bkl,0 ,,
,/,/
iji
kl
bjkbiljbNkibNl
5
Proposed Strategy
B. Band Similarity for Further Bit-Rate Reduction
1. Quantization
/.8nm,0 /
Qλ
ψγ
j8;j0;jfor
i8;i 0;ifor
nm,
nm,ji,nm,
ji,
6
Proposed Strategy
2. Discarding the insignificant bands
it discard and band
ant insignific an as assign ), threshold( If
γ
j)(i,
1-/8image_size
0k
1-/8image_size
0l
ji,k,l
7
Proposed Strategy
. assign error to minimum posses one the
findingby obtained be can matchbest The
γγ
γand γ
for error of valueabsolute theof sum theCalculate
n)(m,
1-/8image_size
0i
1-/8image_size
0j
p,qji,
nm,ji,
,,
(p,q)n)(m,
nmqpD
3. Seeking a best similarity matching band for each significant band
.dby drepresente be can γ
(p,q)n)(m,(p,q)n)(m,n)(m,
8
9
Simulation Results
Fig. 1 Decoded angiogram by the proposed method
Fig. 2 Decoded angiogram by
JPEG
10
(a) (b)
(c)Fig. 3. Test of angiogram image; (a) Original image;(b)Bit rate =0.22 bpp, PSNR=44.98dB;(c)Difference image
11
(a) (b)
Fig. 4. Test of angiogram image; (a) Original image;(b)Bit rate =0.55 bpp, PSNR=38.11dB;(c)Difference image
(c)
12
Simulation Results
Table 1 Performance comparison of JPEG and the proposed method
13
Simulation Results
Fig. 5. Chart of performance comparison
14
Conclusions
• Three major procedures, including band gathering, significance selection and band similarity matching having been presented to reduce the bit rate .
• The proposed method is equal-sized sub-band decomposition and can achieved the perfect reconstruction without quantization.