Fast and robust techniques for detecting straight line segments using local models

Abstract A new line parameterization model, based on Hough transform, offers robust identification of straight lines using geometrical features of lines. The article presents a new algorithm, based on this model, that achieves fast linear (in the number of edge points) detection of line segments, using a fixed small one-dimensional parameter space. The work provides schemes for easy parallel processing and prediction accelaration. The work also suggests a novel contour segmentation technique for filtering out outlier noise, and linking contour edge points at linear time. Results on several real and synthetic images are demonstrated and compared with Random Hough Transform.