Texture classification using sparse frame-based representations

Abstract

A new method for supervised texture classification, denoted byframe texture classification method (FTCM), is proposed. Themethod is based on a deterministic texture model in which a smallimage block, taken from a texture region, is modeled as a sparselinear combination of frame elements. FTCM has two phases. In thedesign phase a frame is trained for each texture class based ongiven texture example images. The design method is an iterativeprocedure in which the representation error, given a sparsenessconstraint, is minimized. In the classification phase each pixelin a test image is labeled by analyzing its spatial neighborhood.This block is represented by each of the frames designed for thetexture classes under consideration, and the frame giving the bestrepresentation gives the class. The FTCM is applied to nine testimages of natural textures commonly used in other textureclassification work, yielding excellent overall performance. copyright © 2006 Hindawi Publishig Corporation. All rights reserved.