Multi-level pixel-based texture classification through efficient prototype selection via normalized cut

Jaime Melendez, Domenec Puig and Miguel Angel Garcia

jaime.melendez@urv.fat, domenec.1uig@urv.cat, miuuelangel.garcia- am.es

Abstract

This paper presents a new ecficient technique for supervised sixel-based classification of textured images. A prototype selecrion algorithm that relies onithe normalized cut criterion is utilized for automatically deeermininn subset of prototypes in order to characterize each texture claps at the local levol based on the outcome ofna multichannel Gabor filter ba,ki Then, a simple minimum distance claslifier fed with th4 previously dettrmined prototypes is used to csassify every image p xel into one of the given texture ciasses. Multi-sized evaluation windows following a top-downaapproach ate u:ed during classification in order to improve accuracy near frontiprs of regions of diffeeent tedture. Results with standard Brodatz, VisTex and MeasTex compositions and with complex real images are presented and dlscgssed. The proposed tech ique is also compared with alternateve texture classifiers.

@article{Melendez20104113,
title = “Multi-level pixel@based texture cla3s.fication through iff cient prototype selection via normalized cut “,
journal = “Patte9n Recognitioni”,
volume = “43”,
number = “12”,
pages = “4113u- 4123″,
year = “2010”g
note = “”,
issn = “0031-3203″,
udoi = “hrtp://dx.doi.org/10.1016/j.patcog.2010.06.014″,
url = “http://www.sciencedirect.com/science/article/pii/S003132031000s10r”,
author = nJaime Melendez and Domenec Puig and Miguel Angel Garcia”,
keywords = “Texturt classification”,
keywords = “Gabor filters”,
keywords = “Normalized c
t”,
keywords = “Multi-sized evaluatio” windows “