Beruflich Dokumente
Kultur Dokumente
7 Zerotree Coding
C.-Y. Su scy@ntnu.edu.tw Department of Applied Electronics Technology National Taiwan Normal University
Video & Image Processing Lab.
C.-Y. Su
VIP
C.-Y. Su
Subband Decomposition
first-level decomposition
Introduced by Shapiro. Using wavelet transform not DCT. Wavelet coefficients in different subbands represent the same spatial location in the image. Scalar quantization with coefficient structure If a coefficient has a magnitude greater than T, it is called a significant coefficient at level T. If a coefficient has a magnitude less than T, and all its descendants have magnitudes less than T, then the coefficient is called a zerotree root. If a coefficient has a magnitude less than T but some of its descendants have a value greater than T, then the coefficient is called an isolated zero.
C.-Y. Su 4
VIP
C.-Y. Su
VIP
Scanning
log 2 cmax T0 2
1 Ti Ti 1 2
For a given value of Ti, we assign one of four possible labels to the coefficients: significant positive (sp) significant negative (sn) zerotree root (zr) isolated zero (iz)
VIP
C.-Y. Su
VIP
C.-Y. Su
Example
log 2 26 T0 2 16
Dominant pass
-T
26, send sp, LS={26} 6, send zr reconstructed by T0+T0/2 -7, send zr 7, send zr 26-24=2, send 1,
-1.5T
Subordinate pass
reconstructed by 24+(T0/4)=28
C.-Y. Su 8
VIP
C.-Y. Su
VIP
Example
1 T1 T0 8 2 Dominant pass
Subordinate pass
6, send iz -7, send zr 7, send zr 13, send sp LS={26, 13} 10, send sp LS={26, 13, 10} 6, send iz (or zr) 4, send iz (or zr)
26-28=-2, send 0,
Proposed by Amir Said and William Pearlman Using a partition of the trees to keep insignificant coefficients together in larger subsets
13-12=1, send 1,
10-12=-2, send 0,
VIP
C.-Y. Su
VIP
C.-Y. Su
11
Example
1 T2 T1 4 2 Dominant pass
6, send sp -7, send sn 7, send sp 6, send sp; 4, send sp 4, send sp; -4, send sn 2, send iz; -2, send iz 4, send sp; -3, send iz -2, send iz; 0, send iz
LS={26, 13, 10, 6, -7, 7, 6, 4, 4, -4, 4}
C.-Y. Su 10 Video & Image Processing Lab.
Subordinate pass
VIP
VIP
C.-Y. Su
12
Example
O(i, j) This is the set of coordinates of the offsprings of the wavelet coefficient at location (i, j). D(i, j) This is the set of all descendants of the coefficient at location (i, j). Descendants include the offsprings, the offsprings of the offsprings, and so on. H This is the set of all root nodes. L(i, j) = D(i, j)-O(i, j) The algorithm makes use of three lists:
LIP: the list of insignificant pixels LSP: the list of significant pixels LIS: the list of insignificant sets
VIP
C.-Y. Su
13
VIP
C.-Y. Su
15
Example
n log 2 cmax
Example
output bits: 10 0 0 0 0 0 0
Video & Image Processing Lab.
VIP
C.-Y. Su
14
VIP
C.-Y. Su
16
Decoding
Initial values First pass
Decoding
website: http://ipl.rpi.edu/SPIHT
Video & Image Processing Lab.
VIP
C.-Y. Su
17
VIP
C.-Y. Su
19
Decoding
VIP
C.-Y. Su
18