COMPARISON OF SAR SEGMENTATION ALGORITHMS
Eduardo K. Viegas, Dalle Lucca, Corina da Costa Freitas, Alejandro C. Frery, Sidnei João Siqueira Sant · 1998
Abstract. This paper aims at comparing the performance of two segmentation algorithms, the MUM (Merge Using Moments) and RWSEG, from simulated data, containing regions with differents homogeneity degrees, for land use aplications. The process for obtaining simulated images consists of criating a phantom (class idealized image) which summarizes the main geometric and topologic characteristics of targets. Then a statistical modelling of observations from each class through a particular distribution is proposed. The performance of the algorithms in study is evaluated from qualitative and quantitative analysis of the acquired results. The quantitative analysis is done from empiric evaluation methods of a segmentation. In order to reduce the influence of particular images on the performance assesment, a Monte Carlo experience is performed.