Adaptive size for operators in line image processing

The authors apply, for each pixel, operators of varying size. These adaptive-size operators can be obtained by selecting, in each location, the size of the operator which gives the best response according to a given criterion. A method is defined for computing a set of Laplacian operators of different sizes from local bicubic approximations. These different Laplacian operators make it possible to obtain a representation of the images, based on a visual model, constituted of a stack of responses. This representation is called the Laplacian multilayer representation. With this method a good segmentation of line images was obtained. An application to the detection of the median lines of the spectrum images is presented.<>