Extended randomized hough transform for 2-D arbitrary shape recognition

The extraction of arbitrary 2-D shapes according to specific templates is a very important operation for object recognition in digital image processing and computer vision fields. Because of its robustness to noises and discontinuity of feature points, Generalized Hough Transform (GHT) is a classical and effective technique to tackle this problem. However, as an extension of the Standard Hough Transform, its computational complexity and memory requirement are considerably large especially when there is no prior knowledge for the orientation and scale of the scene object. The main purpose of this thesis is to develop an alternative algorithm for GHT, to overcome its shortcomings while maintaining its significant advantage of robustness. Based on the idea of Randomized Hough Transform, which is a typical probabilistic Hough Transform, four new algorithms are developed in tins thesis to deal with different cases of object recognition. As extensions to the basic Randomized Hough Transform from analytic curve detection to arbitrary shape detection, the random sampling mechanism, convergent mapping mechanism and dynamic list structured parameter accumulator are used in these proposed algorithms. Compared with the GHT and Template Matching approaches, their computational complexity and memory requirement are reduced while their robustness is retained.Master of Engineerin