Extraction de caractéristiques dans les images par transformée multi-échelle

The aim of this work is to extract features of an image from the result of a multi-scale transform, the Curvelets transform. This one is able in the same time to characterize textures of images according to various scale levels and to take into account the great diversity of shapes and their strong anisotropy even in a disturbed environments (disturbed background, incomplete shapes). Extracted features are curvature and orientation of shapes' contours. Those features are essential in our work. Results are presented in the information retrieval context over ancient manuscripts.