the PAPRI method Do Minh Phuong creator Borne, Frederic Chairperson Honda, Kiyoshi Examination Committee Andrianasolo, Haja Examination Committee Government of Japan Scholarship donor text series technical report th Bangkok Asian Institute of Technology 2001 continuing eng 66 leaves The ultimate aim of any pattern recognition system is to achieve the best possil: classification accuracy for the problem to be solved (J.S. Sanchez, 1997). Classification remote sensing images is usually carried out by using approaches aimed at minimizing t overall error affecting land-cover maps (Lorenzo Bruzzone, 2000). A wide range of patte recognition techniques have appeared during the last two decades for informati extraction from remotely sensed data (Farid Melgan~ 2000). The most popular technique the maximum likelihood classification method known for its good performance a robustness. In addition, an approach to the problem of classification using artificial neu networks has been developed (S.B. Serpico and F. Roli, 1995). The implementation other image classification methods often result into fuzzy image map, where noise arise~ one of the most vital obstacles to a good landscape image. As a matter of fact, m< research conducted on satellite image classification have been focusing on this soluti However, very few applications can partly solve the problem. Recognition of objects extracted from remotely sensed imagery requires matching of object properties with prior stored knowledge. Various properties were use1 this study to form the model of a priori knowledge. The PAPRI (P Aysages defini PRiori - Landscape a priori Defined) method (Frederic Borne, 1994) proposes a good ' for classification of an image that results in a thematic layer with less noise. That met segments an image according to its textural properties as previously defined. It giv1 cutting of the image in texturally homogeneous areas, called "Landscape Units". ' product is nearer of a map than a usual classification as it proposes a synthetic fast global understanding of the image. Although there has been a lot of developmen segmentation of gray tone images in this field and other fields, like robotic vision, t has been little progress in segmentation of color or multi-band imagery. The paper pres an original way for treating a remote sensing image, with the goal to integrate it in a (Geographic Information System). The result of texture analysis is a raster imaE matrix). It is not really ready for use in a GIS. Therefore it is vectorised and a con characterizing attribute is assigned to big units to obtain the "Landscape Units" layer, rather than "Physical Environment" oriented (Frederic Borne, 1994). After ~ topological operations, it will constitute an important element of our information bas1 implement the method, an image processing software has been developed. A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering, School of Advanced Technologies Thesis (M. Eng.) - Asian Institute of Technology, 2001 Remote-sensing images Pattern recognition systems Asian Institute of Technology. http://203.159.5.9/ait-thesis/detail.php?q=B09436 http://203.159.5.9/ait-thesis/detail.php?q=B09436 020208 20260818213314.0