The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
1113
if\i- j |< m
otherwise
(9)
The constant m defines the width of the neighborhood.
In fact from the view of neurobiology the feedback intensity
between the central cell and other neighbor cells has business
with the distance, so neighborhood should be the function of
distance. Considering the characteristics of human visual
system and under our many experiments, this paper introduces
the gaussian neighborhood series kernel function to express the
different topology of TIC A, such as:
1 j-V(i— n ) 2 +(j—m) 2
M w ij, w „m) = e 2
(10)
The new introduced neighborhood functions can obtain the
image basis with obviously enhanced directionality, which has
advantages for the coming image analysis task.
3.2.3 Modified Learning Rule: To resolve the separation
matrix W, the optimization problem can be induced as follows:
ÌÌ
min J(w)= E{G( z(0) 2 )}
M
Il II 2 1
subject to ||wj| 2 = l
(ll)
The Lagrange function can be derived as:
L(w,X) = -E{G(¿h(i,j)(w^z(t)) 2 )} + M||w|! 2 -1) (12>
i=l
Finally the batch learning rule can derived as:
w'— W- Ti(E {g(y)z} - E {g(y)y} w) (i 3)
The information that remote sensing image represent is the
reflectivity of different objects in certain band. Each band of
multi-spectral remote sensing images can be considered as the
combination of reflectivity of the several independent land
objects in certain law. Applying ICA to multi-spectral remote
sensing images, we can obtain the independent component
bands that concentrate the information of specific land objects,
resulting in enhancing the degree of separation of different
objects.
For single band remote sensing image, most important
information such as edge features, texture features are nearly
correlative with high-order statistics. High-order statistics
reflect the important structure and phase feature of image.
Image analysis using ICA/TICA with high-order statistics has
particular advantage, it can realize sparse coding, meanwhile,
ICA/TICA is excellent edge filter (Zeng,2005). When people
observe image, a series image patches are picked up firstly and
then the whole image. Suppose each image patch is denoted by
x, which can be regarded as a linear combination of the base
function matrix A, independent component 5 is the statistic
independent random vector, expressing the coefficients that the
N
corresponding basis act on image, i.e. x = '^a j s i , where
;=l
A = (a,, a 2 , • • •, a N ) ,column vector a t (i = 1,2, • • •, N) denotes
a group of N 2 X 1 pixels basis images. Through ICA resolves
the separation matrix W, one can get the coefficients projected
in independent component basis by y = Wl, which express
the image features in ICA domain. Figured are basis matrix A,
basis vectors have orientation in space domain and localization
in frequency domain, depict most of the edge features of image.
Figure.2 illustrates the basis vectors obtained by our improved
TICA, one can observe the spatial correlation of basis
introduced by topography, the basis offer a more
comprehensive representation compared to the general ICA
model.
Where TJ is learning rate, here the self-adaptive adjustment Figure 1. ICA basis of natural image data
method is developed in this paper.
Through introduction of Lagrange operator to solve the
optimization of TICA, the method has religious deduction
procedure and well property of convergence.
In short, this paper introduces the new topographic kernel
functions to express the relationships between the independent
components, which can better satisfy the human vision system
demand than the former model. Further more, the paper also
gives the new optimization rule to realize the farther
development of TICA. The proposed modified TICA is more
applicable in image fusion.
Figure 2. Improved TICA basis of natural image data