Generalised Hough transform in object recognition

A novel paradigm for recognition of 2D shapes which is based on a modification of the generalized Hough transform (GHT) is presented. The essence of the approach is to confine the choice of shape reference points to the object boundary. In contrast to the classical GHT the memory requirements of the proposed method are a function of the complexity of shape description. This implies that the use of more concise shape representation in terms of higher level features will lead to a further dramatic reduction in the required storage space. Moreover, the burden of increasing the shape representation redundancy to maintain robustness to occlusion and missing data will also be correspondingly reduced. Thus the proposed method, in conjunction with such a feature based representation, can provide a computationally efficient solution to the shape detection problem. Extensive experimental results are presented to verify the theoretically predicted performance gains of the proposed method with the classical one.<>