Fast recognition of lines in digital images without user-supplied parameters
Andreas Busch · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1994
In this paper a method to detect lines in digital images is introduced. The model used to represent the shape of a line is a second order polynomial function of the row and column coordinates. Line positions are delivered with sub-pixel resolution. A threshold required by the method is estimated from the image data, so that the line detection runs off all-automatic without user-supplied control parameters. The robust estimate of the threshold is applicable to edge detection, too. Since directional information is a distinctive property of linear features, the model takes the line direction into account and makes use of it to classify line ends, to close line gaps, and to narrow lines which are wider than one pixel. Additionally, pixels are recognized where lines converge or cross. The performance of the method is shown using SPOT and TM data as examples.