Evaluation on SPOT data of classification algorithms based on Markovian modelization

CNES (French Space Agency) has developed a research program related to SPOT imagery to deal with cartography topics. Many studies, conducted with different laboratories, are intended to work on remote sensing data. The main purpose of the present research is information extraction (network extraction, urban area extraction, segmentation, etc.) One of these studies, made in collaboration with INRIA Sophia Antipolis, intends to classify remote sensing images using MRF (Markov random field) modelization. The paper presents an experiment, conducted by GEOSYS, on crop surveys. An evaluation of the MRF based algorithms is proposed to estimate the results in a supervised context, in order to validate this new approach. A comparison between the proposed methods and standard classification techniques have been done on multispectral SPOT data (XS1, XS2, XS3).