This paper describes method for detection of short line segments in digital images. It aims at identifying buildingsin images taken from the ground view. The process starts with the image edge map and is carried out in twodifferent levels. One is to detect long line segments usually stemming from façade edges and building silhouettes.The other one identifies shorter line segments which typically represent architectural details such as windows andentrances. Selected individual connected components in both vertical and horizontal gradient component mapsare used respectively as input to the Hough transform at this level. Our first result shows that this method iscapable of recognizing lines of interest but has also included many randomly oriented lines. The next step will beto eliminate the random line segments and correlate line segments of the two levels to classify high-level features ofbuildings in an image.