On the detection of structures in noisy pictures

Abstract Hough proposed an algorithm for detecting lines in pictures. His algorithm, based on a slope-intercept parameterization of lines, was improved by Duda and Hart through the use of angle-radius parameterization. When pictures contain random noise that cannot be removed, the Duda-Hart procedure can yield unsatisfactory results. This paper presents two modifications of that procedure which compensate for noise. One method is applicable when the distribution of the noise is known and the other can be used when it is not. The proposed modification is also illustrated for circle detection.