Extracting a new fractal and semi-variance attributes for texture images

Abstract

Texture feature extraction is one of the essential functions in the field of image processing and pattern recognition. There is a very high demand for finding new attributes for this purpose. The fractal dimension (FD) is demonstrated to be an excellent parameter to analyze textures at different scales. In this work, we propose new attributes for image categorization by utilizing two components of texture analysis: fractal and semi-variance characteristics. A set of five attributes is used to investigate different texture patterns. Lacunarity and two other attributes, along with fractal dimension, are four candidates for semi-variance estimation used to ensure a better description of the textured appearance. The simple K-means method was adapted for clustering purposes and revealed from two to ten different clusters. Subsequently, several classification algorithms were used to categorize a new image from the extracted features; these classification algorithms are Nave bays, Decision Tree, and Multilayer Perceptron. The Ten-fold cross-validation scheme is also used to reduce the variability of the results.