Contribution to contours closing by a Markovian approach

Summary form only given. A major problem in edge detection operators is the one of low detection. In some cases, detected contours are fragmented or are not closed. This is because some elements are difficult to extract like corners and multiple contours junctions. To overcome this inconvenience, a basic solution is to connect all neighboring contours segments having the distance between closer extremities inferior to an empirical threshold. This method presents the disadvantage of closing also the contours that mustn't be, even as it can't shut contours that must be and this is in the case where the distance between the segments exceeds the chosen threshold. To overcome this problem, we present a new method for contours closing based on a Markovian modellization. This method transforms the contours closing problem into a labeling one that we solve with an ameliorated simulated annealing. Our contours closing scheme yields good results on both real and synthetic images since most of elements as corners and multiple junctions are clearly closed regarding to the criteria we impose.