Análise de textura através de descritores fractais por meio da utilização de rótulos na transformada de distância 3D

Abstract

This work proposes a method for analysis of color textures based on fractal geometry. The proposal is based in conversion of a two-dimensional image in a 3 dimensional cloud points where each pixel corresponds to a point. After that we set labels for certain groups of points and then dilate this cloud structure. The descriptors are obtained by the volume ratio of components, their labels and dilatation radius. In this way the descriptors can measure both the complexity of the intensity of the pixels models, as well the interference patterns between the labeled components. This combination provides a descriptor robust and precise for texture analysis. The new method is evaluated through statistical analyzes over supervised machine learning problems.