The task of image indexing within semantically meaningful categories remains a very challenging task in the domain of content based image retrieval. The source of this problem lies within the amount of information we draw from each image and the way we extract it, therefore the choice of the descriptors to use is a very important step in the design of a classifier. In this paper we present a city/non-city classifier based on a specifically designed descriptor which is based on a variant of Hough transform called "Fast connective Hough transform" which allows us to deal with line detection efficiently. Here the purpose is to use as few good descriptors as possible to avoid high dimension features vectors which increase computing time dramatically.