The International Archives oj the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
2) Calculating the eigenvalues and eigenvectors according to
the correlation matrix;
6) Replacing the first principal component by the higher
resolution band;
3) Sorting the eigenvalues and eigenvectors;
4) Calculating the principal components one by one according
the PCA transform;
pc = ®-xs L
P an OJ) = P c V,iJ) +S VJ)
(3)
7) Obtaining the fusion results after inverse PCA transform
5) Selecting the first principal component;
xs H =n-pc\ where n = ®'‘ (4)
xs"
~«ll
«12
« 13
« 14
pan
~«n
(Ù\2
«13
«14 _
V
+
XS2
«21
«22
«23
cd 24
pc 2
«21
CO 22
co 23
«24
pc 2
xs"
«31
co 32
OJ 33
«34
pc 3
«31
co 32
co 33
00 34
pc3
1
%
* a:
1
_« 41
«42
OJ 43
«44.
_ P C 4 _
_« 4 >
co 42
0) 43
«44.
PC 4
«12
«13
«14
«11 "
«12
«13
«14 _
r -1
’«11 ~
«22
PC 2
PC 2
• (pc x + 8) +
co 23
«24
PC3
=
«21
•PC, +
«22
«23
«24
pc 3
+
«2.
co 32
«33
«34
PC 4
«31
«32
«33
«34
«31
PC4
_«42
«43
«44.
_«41 _
_« 42
«43
«44.
«41 _
«1.
«12
«13
«14'
PC\
«11"
Xs/
«11'
«21
«22
«23
«24
pc 2
+
«2.
■8 =
xs 2
+
«21
«31
«32
«33
«34
pc 3
«31
xs/
«31
«41
«42
«43
«44.
_P C \ _
.«41.
1
a
1
.«41.
(5)
Thus, the whole algorithm mainly consists of calculating the
correlation matrix, forward PCA transform and inverse PCA
transform. Through deduction the final fusion results are
pan = c, • xs/ + c 2 ■ xs 2 + c 3 • xs/ + c 4 • xs/ + 8
(7)
xs
(k,i,j)
XS
(k,i,j)
CO,
where S = pan-pc t (6)
2) Computing the linear combination of blocks from
multispectral bands in terms of coefficients.
4.2 Modulation-based Fusion Technique
The typical algorithms applying the modulation-based fusion
technique include Brovey, SFIM and HPF fusion algorithms.
To illustrate the deduction for the modulation-based fusion
technique, following is the transformation steps taking
Block-regression fusion algorithm presented by the authors as
an example.
XS
The lower resolution multispectral band k is resampled to
have the same size as the higher resolution panchromatic band
P an after those bands are co-registrated: XSk rs P( xs k ) ?
and after the resampling the implementatiom steps for
Block-regression based fusion algorithm are as follows (Zhang
and Yang,2006):
1) Obtaining linear regression coefficients through multiple
linear regression between the blocks from the panchromatic
band and from the multispectral bands. ^ is equal to 4;
syn = c x • xs/ + c 2 ■ xs/ + c 3 • xs/ + c 4 • xs/
(8)
3) Finishing the fusion operation for every block using tt
following expression:
xs 1 /
xs/
1
%
1
xs/
xs/
xs/
pan
xs/
syn+ 8
XS/
xs/
—
* —
—
xs/
syn
xs 3
syn
xs L 3
H
L
L
XS 4
L x5 4 J
/xs 4 J
L X5 4 j
(1 + —)
syn
xs/
xs/
xs/ /syn
XS2
Ô
XS2
+
XS/ /syn
xs/
syn
xs/
xs/ /syn
1
<3 Tf
%
1
.xs/ _
JCS4 /syn _
(9)