This paper presents a new method of texture modeling using geostatistic theory. From variographics abacus and variogram proprieties, we used fractal and exponential models to characterize Brodatz textures. In this approach, we modeled a texture by a vector called « feature vector » whose components are the parameters characterizing the experimental variogram, notably the « Slope », the « Range », the « Landing » and the « Fractal Dimension ». To estimate these parameters, we use exponential and fractal models. The parameters estimated by this approach help to promote an adequate method of quantitative analysis variogram of textural images. The new method proposed here help also to solve the problem of preferential direction selection often asked by Haralick method of co-occurrence matrix. A comparative study of the proposed method of fractal dimension evaluation and the one proposed in a literature shows that the results obtained are identical with a hundredth of precision on the Brodatz texture images. To demonstrate the applicability of our approach we use to classify a SAR image of ERS-1 from the Atlantic coast of Cameroon. Our approach is one of the great family of supervised classification. It is based on the methods of structural classification. The particularity of this approach lies in the fact that each pixel is fully characterized by its feature vector.
This post was written by Fotsing Janvier (University of Buea). Contact him at firstname.lastname@example.org for more information.