Abstract: This study presents a fast method for road network extraction in satellite images. Our approach comprises two essential phases: the first consists in extracting a curved object; the second phase makes it possible to generate the method of extraction defined in the first phase to be able to extract all the roads. The algorithm of detection consists of three successive stages: Passage of local information to the total information of road (by the detection of the homogeneous elements then linear components of road), optimization with the resolution of the large pixel (a large pixel is a whole of pixels in a window of the image) of which the objective is to obtain a binary vector X for the exact localization of a road at each couple of windows of the treated image and precise detection of the road network (a line determined by the points of passage which adjusts as well as possible on local detections of homogeneous elements of road, these points are detected in an automatic way). Consequently, an improvement of the algorithm defined above consists in applying a threshold to the image of homogeneous elements of road with an aim is to reduce the cost in computing times and to insulate the pixels which one regards as important. Lastly, we propose for the second phase a method based on the principle of the mathematical morphology which is based only on radiometric knowledge of the roads. This developed technique gives good results if the choice of points is made in an optimal manner that wants to say: to minimize the number of points in the linear shape and to maximize these points in the district circular. This algorithm of extraction can be applied to more figures than one will want to combine them and thus all the roads can be extracted.