Straight-Line Detection Using Moment of Inertia

Straight-line detection is a basic step of image analysis for many pattern recognition applications, especially when looking for man made objects. Like points or arcs, straight line segments are primitives on which to base a higher-level representation for object recognition. We propose a novel method for the detection of lines with minimum straightness. It consists of a point linking procedure verifying a straightness constraint measured by the moment of inertia of the segment points. A neighbor point is added to the segment if the enlarged segment has a limited moment of inertia. The implementation was computationally optimized to abandon inappropriate segments as soon as possible and to limit the cost of inertia computation to the current point thanks to an incremental scheme. The straight-line detection algorithm is general purpose and contains an intuitive straightness parameter expressed in pixel. It was applied to several types of images and is demonstrated here for the detection of roads in satellite images in the framework of the EuroSDR challenge and for the detection of conducting stripes of printed circuit board.