Introduction to Genetic Algorithms
Including numerous hands-on problems and solutions, this comprehensive book is a helpful guide and a valuable source of information about Genetic Algorithm concepts for their several practical applications.
TABLE OF CONTENTS Introduction to Genetic Algorithms
10.6: Image Processing
10.6 Image Processing
10.6.1 Designing Texture Filters with Genetic Algorithms
Several techniques have been employed for texture based segmentation. Most of them derive categories of texture descriptors and then, during a training phase, cluster these descriptors to achieve discrimination. Traditional methods of texture feature extraction are based either on statistical or structural models. In the statistical model texture is defined by a characteristic set of relationships between image elements, and for most practical purposes these are determined from tonal values. We will be using one or more of these as comparators, namely
-
Grey level spatial dependency (GLSD) matrices, or co-occurrence matrices and the simplified approach from user based on sum and difference histograms.
-
Texture energy in the spatial domain derived by convolution as described by Laws
-
Methods based on the use of fractals
10.6.1.1 Texture Discrimination Using Genetic Algorithms
In this section, it is to design a mask which, when correlated with the Fourier spectrum of each of the given patterns, will produce a response such that the inter-class difference will be maximized and the intra-class differences will be minimized. Now let's use GA to solve the optimization over all possible masks, by minimizing (or maximizing) an objective function based on the correlation.
Correlation Based Optimization
The steps of the algorithm (Fig. 10.18) can be summarised as:
-
Rectangular patches are selected from a given image as members (training templates) representing each class of texture to be detected.
-
A Fourier Transform (FT) is performed on each of the patches and the...
Copyright Springer-Verlag Berlin Heidelberg 2008 under license agreement with Books24x7
