Biomedical Engineering Reference
In-Depth Information
Chapter 3
Neural Classifiers
In this chapter we shall describe the neural classifiers. One of the important tasks in
micromechanics for process automation is pattern recognition. For this purpose we
developed different neural classifiers. Below, we will describe the Random Thresh-
old Classifier (RTC classifier), Random Subspace Classifier (RSC classifier), and
LIRA classifier (LImited Receptive Area). We will describe the structure and
functions of these classifiers and how we use them. The first problem is the texture
recognition problem.
3.1 RTC and RSC Neural Classifiers for Texture Recognition
The task of classification in recognition systems is a more important issue than
clustering or unsupervised segmentation in a vast majority of applications [
1
].
Texture classification plays an important role in outdoor scene images recognition,
surface visual inspection systems, and so on. Despite its potential importance, there
is no formal definition of texture due to an infinite diversity of texture samples.
There exists a large number of texture analysis methods in the literature.
On the basis of the texture classification, Castano et al. obtained satisfactory
results for real-world image analysis relevant to navigation on cross-country terrain
[
1
]. They had four classes: soil, trees, bushes/grass, and sky. This task was elected
by Pietik¨inen et al. to test their system for texture recognition [
2
]. In this case, a
new database was created. Five texture classes were defined: sky, trees, grass,
roads, and buildings. Due to perceptible changes of illumination, the following
sub-classes were used: trees in the sun, grass in the sun, road in the sun, and
buildings in the sun. They achieved a very good accuracy of 85.43%. In 1991, we
solved a similar task [
3
,
4
]. We worked with five textures (sky, trees/crown, road,
transport means, and post/trunk). The images were taken in the streets of the city.
We took brightness, contrast, and contour orientation histograms as input to our
