This paper describes an automatic process to check the semantic consistency of a segmentation. This process is made possible through the formalism of graphs. In this article we propose to apply this process to the checking of the relevancy of merging criteria used in an adaptive pyramid by matching the obtained segmentation with a conceptual graph describing the objects under consideration. This matching is performed by using a discrete relaxation method. It checks the arc-consistency with bilevel constraints of the chosen semantic graph. The efficiency of this approach is illustrated on synthetic and real images.