A knowledge-based approach to the detection and interpretation of changes in aerial images

<title>Markov random fields and graphs for uncertainty management and symbolic data fusion in an urban scene interpretation</title>

“…The black graph is correct for arc 1,3,4,6 but is wrong for arc 2,6,7. A simple "best of the arcs" (image 3) finds a correct answer where both graphs are correct (1,3,6,7), when the black one is wrong (2,5,7) and when the white one is wrong (4). However the crossroad between (2,4,5) is too extended (radius 9 pixels).…”

“…A simulated annealing (image 4) using the MRF proposed allows the error on this crossroad to punish the choice of arc 4 in the black graph: the result prefers to be wrong on arc 4 than having a large error on this crossroad. Figure 4: all these graphs have been printed in terms of reliability (high reliability in dark, low in bright): in the white graph and the black one, all the crossroads are perfect (Re = 1) ; a simple "best of" provides a too low reliable crossroad (2,4,5) and the improved "best of" gets an overall better result. Each object N13, N1, Bk,, Bk iS the result of the merging of the several data sources, built as before using:…”

“…Several interpretation methods have been designed for the analysis of aerial pictures. Knowledge based systems are often used where the symbolic information is introduced via a rule based system in a declarative way (see for instance [3][4][5][6][7][8][9][10]). Three main aspects of such an interpretation system are interesting to study: the manipulation of the available information (usually using blackboard like systems), the uncertainty management (the key of selfunderstanding and reasoning for the machine) and the last stage, the process of analysis.…”