%0 Conference Proceedings
%4 sid.inpe.br/marte2/2017/10.27.12.38.14
%2 sid.inpe.br/marte2/2017/10.27.12.38.15
%@isbn 978-85-17-00088-1
%F 61623
%T Desenvolvimento de Novas Funções Kernel para Classificação Contextual de Imagens
%D 2017
%A Negri, Rogério Galante,
%@electronicmailaddress rogerio.negri@ict.unesp.br
%E Gherardi, Douglas Francisco Marcolino,
%E Aragão, Luiz Eduardo Oliveira e Cruz de,
%B Simpósio Brasileiro de Sensoriamento Remoto, 18 (SBSR)
%C Santos
%8 28-31 maio 2017
%I Instituto Nacional de Pesquisas Espaciais (INPE)
%J São José dos Campos
%P 2293-2300
%S Anais
%1 Instituto Nacional de Pesquisas Espaciais (INPE)
%X Kernel functions have revolutionized the theory and practice in Pattern Recognition, and consequently the image classification applications. Besides allows the definition of non-linear versions of methods like Support Vector Machine (SVM), such functions allow generalize the application of these methods on the classification of non-vector patterns, such as probability distributions, information sets, etc. This possibility motivates the development of kernel functions able to deal with the context which the pixels are inserted and consequently inducing contextual classifications when adopted. In this initial study, three kernel functions are proposed for contextual classification. These functions are based on stochastic distances, non-parametric statistical tests and spatial variation modeling. A case study about the land use and land cover classification with an ALOS PALSAR image is carried in order to compare the performance of the SVM method though the use of the developed kernel functions. Comparisons with other contextual methods based on SVM are included in this analyzes. The results shows potential on the new proposals, especially the kernels based on stochastic distance and nonparametric statistical test.
%9 Processamento de imagens
%@language pt
%3 61623.pdf