A parallel morphological implementation of the Hough transform

A new parallel architecture is presented consisting of a toroidal memory and a processing element for each of its rows and columns. The standard Hough transform for line detection in digital images and an alternative approach to line detection based on mathematical morphology are implemented on this architecture. In the alternative approach the M line angles to be detected are represented by M structuring elements, which are used to erode the image and hence reduce the average number of accumulator accesses per pixel from O(M) to O(1) for reasonable images. The choice of structuring elements allows this method to be tailored towards a more detailed image analysis than the standard Hough transform. While neither implementation is asymptotically optimal it is argued that for realistic images they outperform, by several orders of magnitude, any of the alternatives presented in the literature.<
>