| 摘要 |
The device for the automatic segmentation of images according to the invention is comprised of a data processor associated with a network of automatons organized in three levels; the inputs of each of the automatons are weighted by adjustable coefficients; the number of automatons of the first level is equal to the number of characteristic values calculated from observation windows taken first from an assembly of examples of image areas having different textures: their apprenticeship by the network allows to adjust the weighting coefficients of the network automatons till all examples of one same texture lead to a same configuration, characteristic of said texture, at the network output. The image segmentation comprises, from the assemblies of characteristic values calculated on image areas to be analyzed, the analysis of the network output in order to deduce the area texture. Application to the segmentation of aerial images or medical images, for example, and more generally to any type of images wherein the regions are discriminatable by their texture. |
