Analysis of Weighted Index for Segmentation Evaluation

The use of region based or object oriented techniques can enhance the accuracy of remote sensing based land coverclassifi cation. Due to the great amount of available segmenters and the need to select its parameters, a methodologyfor segmentation evaluation and comparison is of great relevance. The...

ver descrição completa

Detalhes bibliográficos
Autores: Reis, Mariane Souza, Pantaleão, Eliana, Sant'Anna, Sidnei João Siqueira, Dutra, Luciano Vieira
Formato: artículo
Estado:Versión publicada
Fecha de publicación:2019
País:Brasil
Recursos:Universidade Federal de Uberlândia (UFU)
Repositorio:Revista brasileira de cartografia - RBC (Online)
Idioma:portugués
OAI Identifier:oai:ojs.www.seer.ufu.br:article/49191
Acesso em linha:https://seer.ufu.br/index.php/revistabrasileiracartografia/article/view/49191
Access Level:acceso abierto
Palavra-chave:índice de segmentação
avaliação de segmentação
comparação de segmentadores
segmentação de dados óticos
segmentation index
segmentation evaluation
segmenters comparison
optical data segmentation
Descrição
Resumo:The use of region based or object oriented techniques can enhance the accuracy of remote sensing based land coverclassifi cation. Due to the great amount of available segmenters and the need to select its parameters, a methodologyfor segmentation evaluation and comparison is of great relevance. The Weighted Index for Segmentation Evaluation(WISE) and some variations are analyzed in this work, by comparing them to other two indices from literature, foroptimal segmentation of synthetic images. The indices were also evaluated by the classifi cation results obtained with Neatest Neighbor classifi er. These indices were tested in three diff erent scenarios, that is, considering sets ofland cover classes with diff erent levels of separability. When compared to other indices, WISE and its variationsshowed good results for segmentation selection, while some of the operational and theoretical problems found inother indices were not observed.