Hough transform for line detection using segment voting weighted by surround suppression

Line detection is an important problem in fields of image processing and computer vision. Hough transform and derived methods have been used broadly to solve the problem. When applied on real-world images, these methods often suffer unsatisfactory detection performance from large amount of noise edge points produced by complex background or texture regions. A new Hough transform algorithm for line detection was proposed. By introducing surround suppression measures into the method, impacts of the mentioned noise edge points are reduced significantly. An efficient computation of the surround suppression was given. Furthermore, by using a line-segment-based voting scheme assisted by a fast point cluster detection and partitioning procedure, the voting process is considerably accelerated. The overall algorithm achieved high detection rate while ran at relatively fast speed in experiments. Results show that the proposed algorithm is effective and preferable.