Apport du réseau de neurones pour une détection de contours par transformée en ondelettes

The method exposed in this paper represents a new edge-detection tool of a greylevel image by the cooperation of two technics : wavelet decomposition and neural networks. The first part recalls the necessary background on mono and bidimensional wavelet decomposition and their main properties. The difficult phase of the algorithm lies in the optimal recomposition of different resolutions, in the aim to obtain thin and noiseless edges. This work is given to a neural network which constitutes the object of the second part The main interest of this new method is to give good results with images whose caracteristics are completly different, without to modify any parameters