<title>High-order curve recognition with the Hough transform using a connectionist approach</title>

Abstract

Extraction of lines and curves from images is one of the most important and fundamental tasks in machine inspection and computer vision in general. Among all the techniques in detecting lines and curves, the Hough Transform (HT) method is unique in its ability to cope effectively with noise, gaps in outlines and even partial occlusion. In spite of this ability, the HT method is still not widely used in real time applications due to its computationally intensive requirements. One solution to this problem is to find an architecture for parallel processing. Recently, some approaches using parallel architectures have been reported. In real-time applications, the overall time required using these approaches is of the order of hundreds of milliseconds for a typical image of 256 X 256 resolution. Clearly, this speed is not good enough for most image processing and machine vision tasks where line detection is just partial work. Since these architectures use commercial components and are not highly parallel, there is ample opportunity to improve the speed by using neural-like analog circuitry. An original method of higher order curve (HOC) detection using the Hough Transform is presented. This method is computationally very efficient and may yield to hardware implementation, thus making it possible to use the Hough Transform in fast real time applications.