Circle detection using Hough transform filters

Abstract

A novel formulation of the circle Hough transform (CHT) is introduced which enables the detection of circles of different radii to be made using a single parameter space. We formulate the CHT as one or more convolution operations. Comparisons are made between conventional CHT techniques, formulated as filters, and a novel complex vector CHT. This technique exploits both edge orientation and distance from an edge point coded as complex phase. We present results that show that the novel phase and orientation CHT has a positional accuracy of approximately +or-2 pixels at good signal to noise ratios (SNR) down to approximately -16 dB. Using orientation alone, the positional accuracy is poorer falling to approximately +or-10 pixels but at SNR levels of approximately -25 dB. >