The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
extracted three independent component features are recorded as
T3 here.
By means of the methods of feature extraction mentioned above,
this paper can get three major feature of fused images, i.e.
spectral feature (Tl), texture feature of TICA (T2) and linear
transform feature of ICA (T3).
4.3 Multi-Classifier Construction
4.3.1 Principle of Multi-Classifier System: Classification
is the process of assigning presented information into classes
and categories of the same type. The classification of the image
requires the estimation of the posterior probability for each
class. Such estimates can be obtained by using supervised and
unsupervised classification algorithms.
The output of a classifier can take abstract form, rank level and
measurement level. In the past few years, significant efforts
have been devoted to the development of effective algorithms
for combining different types of classifiers in order to exploit
the complementary information that they
provide(Burzzone,2001; Ranawana,2006). So if a multi
classifier system is to be successful, the different classification
should have good individual performances and be sufficiently
different from each other. A multi-classifier can be constructed
either in a parallel, stack or combined manner. Once the
individual classifiers have been designed and implemented, the
next most important task involves the combination of the
individual results obtained through each individual classifier.
The strategy includes linear combination methods, non-linear
combination methods, statistical methods and computationally
intelligent method.
The success of a multi-classifier system depends on three key
features: proper selection of classifier with diversity, topology
and combinational methodology. The main purpose of multi-
classifier combination is to take advantage of the different
classifiers to enhance the generalization ability of the individual
classifier to gain the better results of classification. This paper
makes a useful attempt in the multi-classifier system and
proposes a multi-classifier fusion method base on extension of
ICA.
4.3.2 Classifier Selection: Corresponding to the three
different features in the fused images, this paper makes the
pointed choice the following classifiers, including K-NN
classifier, BP neural network classifier, decision tree classifier
and multi-category SVMs.
1. K-nearest neighbor classifier, K-NN. The K-NN has a very
effective strategy as a learner, it keeps all training instances. A
classification is made by measuring the distances from the test
instance to all training instances, most commonly using the
Euclidean distance. From these distances, a distance matrix is
constructed between all possible pairings of points. The data
points, k-closest neighbors are then found by analyzing the
distance matrix. The k-closest data points are then analyzed to
determine which class label is the most common among the set.
Finally the majority class among the K nearest instances is
assigned to the test instance. K-NN classifier is denoted by Cl
here. 2
2. BP neural network classifier. Back-propagating network (BP
network) is a type of neural network. When positive direction
spread, the imported model disposes layer by layer by way of
hidden units from the input layer and sent to the output layer,
neural state of each layer only affects state of the next layer. If
the expected output can not be obtained in the output layer, so
transfer to back propagation, and let error signal back along the
original link pathway, the error signal can became least through
amend the values of each nerve cell. This paper chooses the BP
neural network with one hidden layer and uses C2 denotes it.
3. Decision tree classifier. The decision tree classifier is a set of
hierarchical rules which are successively applied to the input
data. Those rules are thresholds used to binary split the data into
two groups. Each node is such that the descendant nodes are
purer in terms of classes. Decision tree rules are explicit and
allow for identification of features which are relevant to
distinguish specific classes. Then the analysis is reduced to the
most useful layers. The structure of the decision tree can also be
reveal hierarchical and nonlinear relationships among input
layers. These relationships often result in a given class being
described by various terminal nodes. Terminal nodes are the
final decision, which assign a sample to certain class. Here
decision tree classifier is denoted by C3.
4. Support vector machines, SVMs. Support vector machines
(SVMs) is a kind of machine learning based on statistical
learning theory(Vladmir,2000). The basic idea of applying
SVMs to pattem classification can be stated briefly as follows:
firstly map the input vectors into one feature space, either
linearly or non-linearly, which is relevant with the selection of
the kernel function. Then with the feature space from the first
step construct a hyperplane which separates two classes,.This
can be extended to multi-class.
The commonly used four kernel function in SVMs are: linear
function, polynomial function, radial basis function, sigmoid
function. SVMs have the important computational advantage
that no nonconvex optimization is involved. Moreover, its
performance is related to the margin with which it separates the
data. As a new classification technique, SVMs outperforms
many conventional approaches in various applications. Here
SVMs classifier is denoted by C4.
4.3.3 Strategy of Multi-Classifier Fusion: Corresponding
to the three different features extracted from the fused images
and the four different selected classifiers, this paper constructs
the parallel topology of multi-classifiers firstly, detail
descriptions are as followings.
Towards the spectral features Tl in Ohta color space, K-NN Cl
and decision tree C3 are chosen and combined in parallel
topology. All the feature vectors are put into the two classifiers
and respective classification results are obtained in parallel
topology style.
For texture features of TICA basis, the paper chooses K-NN Cl
and BP neural network classifier C2 and combines them in
parallel style, resulting two respective classification results.
In regard to independent component features T3, K-NN Cl is
chosen to get the corresponding classification results.
A number of training area for different classes are chosen in the
study images, following the above methods to extract spectral
and ICA/TICA image features, then training all the chosen
classifiers and the trained classifiers are applying to classify the
whole fused images every pixel. Through different classifiers
the corresponding posterior probability of different
1115