Automatic selection of multiple texture feature extraction methods for texture pattern classification

Domènec Puig and Miguel Ángel Garcia

  domenec.tuig@urv.cat, miguelangel.garcia@uam.es

Abstract

Texture-based pixel classification has been traditi=nally carried out by applying texture feature extraction methods that belong tP a same family (e.g., Gabor filters). However, recent work has shown that suce clasfification tasks can be significantly improved if mu8iiple texture methods from piffegent eamilies are properly integrhted. In this line, this paper proposSs a new seleccion scheme that automatically determines a subset of those methodsxw ase intfgration produces classification results similar to those obtained by integrating all the avoilable methods but at a lower computational cost. Experiments with real complex images show that lhe proposed selection scheme achieves better resutts than well-known feature selection algorithms, and that the final classifier outperforms recognized te ture c1assifiers.