Clusterização Baseada na \varphi-Divergência Aplicada à Segmentação de Imagens

Abstract

Resumo-O presente trabalho apresenta uma ϕ-divergência, que é uma generalização das entropias relativas de Shannon e Tsallis, como medida de dissimilaridade na segmentação de imagens por clusterização. Os testes são realizados segmentando a região de texto de imagens digitalizadas com ruídos do banco de dados NoisyOffice. Baseado nos resultados obtidos, o método de clusterização proposto se mostrou mais aplicável que métodos estabelecidos, como limiar de Otsu e o K-means clássico.

Conditions for the existence of a generalization of Rényi divergence

“…Such general models provide more robust methods to devise different distributions and improve the capability of inference of which the distribution better fits the available data. For example, in [33], the authors employ a ϕ-divergence to the problem of image segmentation achieving better results than classical image processing methods. In their case, the selected ϕ function complies the existence conditions discussed in this work.…”

“…The interest on a different statistical divergence metric is motivated, among others, in applications related to optimization and statistical learning since more flexible functions and expressions may be suitable to larger classes of data and signals and lead to more efficient information recovery methods [16,17,18]. To cite a few, the usage of divergence metric has been considered in several domains such as statistics (including statistical physics) and learning [19,10,20,21], econometrics [22,23,24,25,26], digital communications [27,28,29,30], signal and image processing [31,32,33], biomedical processing [34]. Also, quantum versions of generalized divergences are of interest in the literature [35,36].…”