Agrandissement d'images par synthèse de similarités et par induction sur un ensemble
Didier Calle · HAL (Le Centre pour la Communication Scientifique Directe) · 1999
This thesis deals with the enlargement of still digital images in a general context. It is made of three parts. The first one is a detailed description of the enlargement methods which can be found in the literature. First of all, we present classic interpolation methods which aim at retaining the frequencies of the image to be enlarged. Then, we detail more recent methods which retain the structures of images and give better results. The second part is the main contribution of this work, as it presents two original enlargement methods. The first one synthesizes similarities detected on a pyramidal representation of the image, and it modifies and improves both the analysis and the synthesis phases of the classical fractal zoom. In this part, the assumption of the similarities retaining is experimentally checked. The second method which is here presented deals with the admissible set of enlarged images of an original image. This admissibility is closely linked to the reduction of the enlarged image, i.e, an image belongs to the solutions set only if its reduction is identical to the original image. We study different algorithms to project onto this set. In the third part, we present several ways of improving and applying our methods. First, we improve the quality of an image enlarged thanks to the similarities synthesis method : to do so, we detect these similarities on a quincunx pyramid. Then, we use the induction enlarging method to regularize in reduction the images enlarged with the similarities synthesis method. We eventually use the induction enlarging method to obtain a hierarchical encoding of the image which allows us to make its progressive transmission.