Jean-Michel Benharrosh

• Abstract : Very high resolution images, such as airborne images, are full of very interesting details for scene analysis. However, generally, full resolution is not required on a whole scene and only a few areas of interest, as well as structuring cartographic features need it. Thus, having a multiresolution image with some parts at full resolution and others at coarser resolution is sufficient for many applications. Moreover, such images offer some interesting characteristics. Such images, combining different resolutions can be obtained by adaptive filtering of images at high resolution. We propose three methods of adaptive filtering that smooth an image while preserving cartographic structuring features or interesting regions of the image. The first one relies on an anisotropic diffusion method introduced by Saint-Marc, Chen and Medioni. The second is based on wavelet theory. Eventually, the third method is based on an analogy between the image and an electrical network. Each node of the network (i.e. pixel) is connected to his neighbours by means of resistors. Some resistors are representative of the diffusion and others of the importance of the original data in theiffusion. In that purpose, we study the relations between parameters of

the network and the resolution of the filtered image and we demonstrate the links between electrical analogy and a Markov s approach. We estimate and compare these methods underlining advantages and limits of each one. From these methods, we propose two applications : the generation of an adaptive quick-look and adaptive compression. Concerning the first application, we propose a method to generate a subsampled image at low resolution preserving cartographic features designed as areas of interest by user. Concerning the second application, our goal is to transmit optimally on a low rate network an image preserving some cartographic features at high resolution in a low resolution environment. Thus, adaptive compression links some classical methods of compression with multiresolution techniques. All these theoretical results are experimentally checked and some images of diffusion, quick-look and compression are shown in this manuscript.

• Keywords : Multiresolution analysis, Anisotropic diffusion, Quick-Look, Adaptive and selective compression.