New parallel Hough transform for circles
The Hough transform is a well known medium-level image recognition technique for the detection of curves. The conventional Hough technique [1, 2] requires a three-dimensional accumulator array (AA) for the detection of circles. Because shape parameterisation and data structure of the accumulator array significantly affect the memory space and computation loading requirements of any hardware or software realisation, they should be carefully selected for any fast and efficient algorithm. A new scheme which uses only a pair of two-dimensional accumulator arrays to reduce the storage and computation time by an order of magnitude or more is proposed. This new scheme is capable of discriminating multiple (including concentric) circles in a complex real life image with a recognition rate of 95–100%. Various parallel realisations of this Hough scheme for circles on a general purpose MIMD (a reconfigurable transputer network) machine are discussed and a comparison of their performances with the conventional approach on the basis of execution time and recognition rate is presented.