Texture description using fractal and energy features

Abstract

The fractal dimension of a texture has been used in the past as a segmentation feature, but since it cannot sufficiently describe enough textural characteristics, additional features are needed. In this paper we demonstrate that by combining the fractal dimension with a simple textural energy measure, a significant performance improvement is achieved compared to using each feature alone. The fractal dimension is computed using an efficient method that is also more accurate than most other popular methods, and the textural energy is easily computed using convolutional masks. Segmentation and classification of natural textures based on these two features is presented and the effect of additive noise is considered.

Automated fabric defect detection—A review

“…In addition, the method in [65] achieved an accuracy of 93.85% for classifying 65 anonymous texture samples from the Brodatz texture database [67]. There are two further publications [68,69] referring this technique. Finally, in a comparison of this technique and another fractal method [68], the technique in [65] was found to be computational demanding even in noise-free texture samples.…”

Classification of noisy signals using fuzzy ARTMAP neural networks

“…First, we define the distance between a template and a pattern as (16) In the special case where the template is a pattern , the distance between and as it is defined in (16) is equal to (17) Fig. 4 presents some examples of the distance as it is defined by (16) and (17).…”

“…The features that are selected must be sufficient to characterize the texture. Examples of features that have been used in the past include Gabor energy [8]- [10], Fourier transform energy [11], second-order statistical features [12], wavelet features [13], [14], and fractal dimension (FD) [15]- [17].…”

Multiscale image segmentation using fractal and neural network

“…Peli [ 12] showed that natural texture is of fractal feature and man-made objects are not of it, so fractal dimensions can be used to distinguish man-made objects from natural background. Dubuisson [ 13] and Kasparis [14] concluded that fractal dimensions at only one scale, as segmentation feature parameters, are not sufficient to segment natural texture images. To overcome this problem, two kinds of methods are applied.…”