block truncation coding using pattern fitting
DESCRIPTION
Block Truncation Coding Using Pattern Fitting. Source: Pattern Recognition, vol.37, 2004, pp.2131-2139 Authors: Bibhas Chandra Dhara, Bhabatosh Chanda Speaker: Shu-Fen Chiou( 邱淑芬 ) Date: 2004/09/08. Outline. Introduction BTC (Block Truncation Coding) Purposed method - PowerPoint PPT PresentationTRANSCRIPT
1
Block Truncation Coding Using Pattern Fitting
Source: Pattern Recognition, vol.37, 2004, pp.2131-2139Authors: Bibhas Chandra Dhara, Bhabatosh ChandaSpeaker: Shu-Fen Chiou( 邱淑芬 )Date: 2004/09/08
2
Outline Introduction BTC (Block Truncation Coding) Purposed method Experimental results Conclusions Comments
3
Introduction Since 1970s the block truncation coding
(BTC) has been studied a lot.•High PSNR but low compression ration.
A modified scheme of BTC is proposed to achieve high compression ration and keeping PSNR as high as possible.
4
BTC (1/3)
將影像分成不重疊的n n方塊
計算每個方塊的量化臨界值與重建階
建立位元圖 編碼並送出位元圖
編碼並送出重建階
原影像
解碼位元圖 重建影像方塊 組成重建影像
編碼端
解碼端
建立重建階
5
BTC (2/3)14
614
915
215
6
9712
214
414
7
89 9013
514
5
85 92 9912
0
12316/)120152149146( X
1478/)145152149146( b
44 block
998/)1208912297( a
146
149
152
156
97122
144
147
89 90135
145
85 92 99120
146
149
152
156
97122
144
147
89 90135
145
85 92 99120
1 1 1 1
0 0 1 1
0 0 1 1
0 0 0 0
ba , +
147 147 147 147
99 99 147 147
99 99 147 147
99 99 99 99
X
X
Bit pattern
216
1688
bpp
6
BTC (3/3)
(a) Original image (b) Reconstructed image with conventional BTC(PSNR=32.89 and bpp=2)
7
Purposed method
將影像分成不重疊的n n 方塊
Pattern fitting
Encoding the block
Entropy coding and send
原影像
receive 重建影像方塊 組成重建影像
編碼端
解碼端
8
Pattern fitting (1/4)
Set of two-level patterns
Njp j , 2, ,1 ,
Pattern book
9
Pattern fitting (2/4)146
149
152
156
97122
144
147
89 90135
145
85 92 99120
1p
2p
0
01 2
0
ji pxj ike xf
1
11 2
1
ji pxj ikke xf
1
0
/1
and /1
where
1
0
ji
ji
pxi
pxi
xfkk
xfk
10 jjj eee
jj em min arg
44 block
408318901 e
1450 1001
1390 1061
8204673532 e
1,emin 21 em j
8 ,16 kk
8 ,16 kk
10
212
2
1
22
11
1
1
1
)(2
2
2
mm
xfk
m
xfk
m
kkk
kd
kkk
kkmA
dAb
dAa
k
ii
k
ii
146
149
152
156
97122
144
147
89 90135
145
85 92 99120
44 block
2p
1m
8 ,16 kk
26)816(82
1626
12381682
)1682(26123
26
15797
123
2
1
d
A
m
m
Pattern fitting (3/4)
11
Pattern fitting (4/4)
Output image after pattern fitting
79.31PSNR
375.116
688bpp
664log
64
, 2, ,1 ,
2
N
Njp j
12
Encoding the block (1/2)
otherwise,
, if 0
th
th
dd
ddd
88.0bpp
61.31PSNR
4
thd
Considering entropy coding for and 8-bits for Ad
13
Encoding the block (2/2)
minmax2
minmax
min
log exceedingnot interger largest theis where
2
AAl
AA
AAA
l
4
1iiic AwAA
0.64bpp 31.57,PSNR
codingentropy with ,
dA
14
Reconstruction
,0 if
,0 if 0~
2
~min
minmax
ddd
dd
AAAA
A
th
l
15
Experimental results
Original image Reconstructed image
16
Experimental results
Original image Reconstructed image
17
Experimental results
Conventional BTC (PSNR=32.89 and bpp=2) Purposed method (PSNR=31.59 and bpp=0.64)
18
Experimental results
Experimental result of proposed method and others
19
Conclusions Instead of determining bit-pattern for a block, an
optimum bit-pattern is selected from a pattern-book.
Our method is presented to achieve higher compression ration and keeping PSNR as high as possible.
20
Comments 如何決定 wi? 計算壓縮率時 , 是否也應考慮 Pattern book?