ABSTRACT
Recognition of textured objects is one of the main challenges for vision-based measurements and control systems. One of the main problem is choosing properly defined features that would extract image objects from the background. The texture features may be characterized by their statistical, structural and dynamic properties. Statistical based methods use both parametric and nonparametric models of the image textures such as Bayes analysis, Hidden Markov Models, Regression, Linear Autoregression (AR), Nonlinear Autoregression (NAR), and Autoregression with Moving Average (ARMA). The eigen-harmonic decomposition (EHD) is based on planar and spherical functions that have been used for compact presentation of textured images as well as human facial images in some articles. Texture modeling and recognition with properties of invariance to space and scale perturbation can be made by including transforms of initial image into its invariant projections as surface curvature and Beltrami flow.
