A framework for designing optimal Hough transform implementations

This paper develops the theory underlying the generalised Hough transform (GHT), with the aim of minimising computational load. The approach adopted aims to offset the extra load resulting from use of complex features by the reduced PSF voting dimensionality they incur. Calculations show that naive application of this idea results in a worst-case solution, but further studies reveal a number of systematic means of overcoming the problem. Of particular importance is the need to skim past unsuitable complex features, and to limit the number of features employed, to avoid diminishing returns in the accuracy achieved. Overall, the paper provides a unified view of means for reducing the computational load of the GHT