الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 96 |  |  |
| | | from | 96 | | |
Abstract
Texture is an important cue allowing humans to discriminate objects. It contains important information about the structural arrangement of surfaces and their relationship to the surroundings so it is an important property of images.
Therefore, characterizing image patterns with textural features, for what is known as texture-based classification, is used in a wide range of applications.
As an attempt to solve the texture-based classification problem, many studies have been done. The challenge that confronted all the developed methods was to overcome the problems of the texture dependency on scale and image geometry. The researchers tried to find an efficient approach for texture analysis that provides a proper description and chara As a result of the deficit in the proposed models in defining relations among the texture representative features, the Genetic Programming (GP) is suggested as an evolutionary algorithm for generating decision trees. These decision trees define dynamic rules that reflect the relation among the features. Moreover, a combination between GP and any other machine learning algorithm is suggested to improve the quality of characterization in texture-based problems.
In the classification phase, the probability density function in the PNN is estimated based on the Gaussian distribution. So, it is suggested to estimate the probability density function using reliability theory. In addition to that, non-parametric approaches are not covered in the proposed work so; unsupervised learning like regression is suggested.
cterization to image in addition to find an appropriate classifier