author = {Trias-Sanz, Roger}, title = { An Edge-based Method for Registering a Graph onto a.pdf

An Edge-based Method for Registering a Graph onto an Image with Application to Cadastre Registration Roger Trias-Sanz March 2004 Rapport technique 2004/1 du laboratoire SIP, CRIP5 Université de Paris 5 avec la collaboration de l’Institut Géographique National BIBTEX document entry: @TechReport{TriasSanz:PhD-RR20041-ICPR04, author = {Trias-Sanz, Roger}, title = { An Edge-based Method for Registering a Graph onto an Image with Application to Cadastre Registration}, institution = {SIP-CRIP5 (Universit{\’e} de Paris 5) and Institut G{\’e}ographique National}, year = 2004, type = {technical report}, number = {2004/1}, address = {45 rue des Saints-P{\‘e}res, Paris, France}, month = mar } Abstract In the context of the development of a land use analysis system, we need to register cadastre edges onto color and near-infrared images at 50cm resolution. The mis-registration to be corrected exists not only because of geometrical deformation or acquisition errors, but because land use does not always follow cadastre divisions. We formalize this as a graph matching problem which can be solved by simulated annealing. As a side effect a score for each cadastre edge is obtained, which shows which edges can be found in the image and which cannot. 1. Introduction Salade is an ongoing project which aims at segmenting high-resolution (50cm) images into fields and other regions with very high reliability. Cadastre data helps by providing additional information about crop distribution and position. However, cadastre and image edges rarely match exactly, because land owners do not usually follow cadastre limits when growing their crops. We present a method for registering a cadastre graph onto an image. Registration of cadastres to images has been viewed as a non-rigid registration problem [2, 6]. The goal is to find the transformation within a class that best converts an image to a reference image. This approach works when the initial mis-registration is due to sensor deformation. In our