Scale Space Texture Classification Using Combined Classifiers

Since texture is scale dependent, multi-scale techniques are quite usefulfor texture classification. Scale-space theory introduces multi-scale differentialoperators. In this paper, the N-jet of derivatives up to the second order atdifferent scales is calculated for the textures in Brodatz album to generate thetextures in multiple scales. After some preprocessing and feature extraction usingprincipal component analysis (PCA), instead of combining features obtainedfrom different scales/derivatives to construct a combined feature space,the features are fed into a two-stage combined classifier for classification. Thelearning curves are used to evaluate the performance of the proposed textureclassification system. The results show that this new approach can significantlyimprove the performance of the classification especially for small training setsize. Further, comparison between combined feature space and combined classifiersshows the superiority of the latter in terms of performance and computationcomplexity.