Image Processing Applications Based on Texture and Fractal Analysis

Image Processing Applications Based on Texture and Fractal Analysis

Radu Dobrescu (Politehnica University of Bucharest, Romania) and Dan Popescu (Politehnica University of Bucharest, Romania)
Copyright: © 2013 |Pages: 25
DOI: 10.4018/978-1-4666-3994-2.ch014
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

Texture analysis research attempts to solve two important kinds of problems: texture segmentation and texture classification. In some applications, textured image segmentation can be solved by classification of small regions obtained from image partition. Two classes of features are proposed in the decision theoretic recognition problem for textured image classification. The first class derives from the mean co-occurrence matrices: contrast, energy, entropy, homogeneity, and variance. The second class is based on fractal dimension and is derived from a box-counting algorithm. For the purpose of increasing texture classification performance, the notions “mean co-occurrence matrix” and “effective fractal dimension” are introduced and utilized. Some applications of the texture and fractal analyses are presented: road analysis for moving objective, defect detection in textured surfaces, malignant tumour detection, remote land classification, and content based image retrieval. The results confirm the efficiency of the proposed methods and algorithms.
Chapter Preview
Top

Background

Texture Analysis Features

Texture analysis has been studied using various approaches, such as statistical (grey level co-occurrence matrices and the features extracted from them, autocorrelation based features and power spectrum), fractal (box counting fractal dimension) and structural (the texture is composed of primitives and is produced by the placement of these primitives according to certain placement rules). The structural approach is suitable for analyzing textures where there is much regularity in the placement of texture elements. The statistical approach utilises features to characterise the stochastic properties of the distribution of grey levels in the image.

There are two important kinds of problems that texture analysis research attempts to solve: texture segmentation and texture classification. The process called texture segmentation involves identifying regions with similar texture and separating regions with different texture. This implies prior knowledge of the texture types and component number which exist in the analyzed image. Texture segmentation is a more difficult problem than the texture classification. The later involves deciding what texture class an observed image belongs to. Thus, one needs to have an a priori knowledge of the classes to be recognized. Because texture has many different dimensions and characteristics there is no single method of texture representation that is adequate everywhere. The features implied in the classification process can differ from one application to another. Wagner (Jähne, 1999) compares the performance of 318 textural features from 18 feature sets. On the other hand, texture classification can be used to segment multi-textured images. With this end in view, the image is divided into small textured region which are classified and indexed. Generally, the result is a coarse type of segmentation process. The segmentation fineness depends on the degree of partition of the initial images. If the degree of partition is too fine, then it is possible that the texture could disappear. Actually, for an application, a partition index is established taking into account the given image resolution and the texture fineness.

Complete Chapter List

Search this Book:
Reset