Texture Classification by Functional Analysis of Size Distributions

I. Epifanio, J. Domingo, and G. Ayala(Spain)


texture analysis, functional data analysis, size distributions, mathematical morphology, curve discrimination.


This paper is mainly concerned with texture classification. Classical descriptors used in texture classification are granulometric size distributions [1]. They are a particular case of the spatial size distributions introduced in [2], which gave very good results in texture classification. As these size distributions are functions, in order to improve the classification in these cases, we propose to use the recent functional data analysis techniques [3], instead of the classical use of raw data (with selection procedure) or moments of these distributions. A typical experimental set up has been performed on a small texture database. The results show the superior performance obtained by using functional discrimination, in concrete with the penalized discriminant analysis introduced by [4], and ICA coefficients [5].

Important Links:

Go Back