Remotely sensed images are often too complex to be analyzed by a single algorithm. In this work, the author considers the problem of merging results issued from different algorithms performing the same task on a satellite image so as to improve the quality of the result. Two types of error measures are computed for each (intermediate) result to estimate its reliability: the global error measure determines whether a result is good enough to be used in the fusion process; the local error measure of each site determines hour information given by this site will be taken into account in the fusion process: site with a smaller local error measure has a higher reliability, thus has more influence on decision making in the fusion process. Map knowledge is used for evaluating the error measures