Grouping Straight Line Segments in Real Images

In this paper, we discuss straight line extraction as a part of the image interpretation process. Favoring the use of line drawings as intermediate data for the extraction, we survey the current methods, which all achieve a polygonal approximation of lines, and show that they are not appropriate for the identification of straight elements in a scene. We propose a new approach which uses a scale invariant criterion and is based on the characterization of prime segments in a line, and develop an original method for obtaining these prime segments. Results show that we significantly improve the performance of straight line extraction. The methodology we have used here is applicable to a large class of segmentation problems.