Wavelet Based Affine Invariant Representation for Object Recognition
Quang Minh Tieng, Wageeh Boles · International Symposium on Information Theory and its Applications · 1994
A novel algorithm for representing and recognising planar objects undergoing a general affine transformation is presented. The proposed algorithm, which is based on the dyadic wavelet transform, makes use of the approximation and detail signals of the object contour coordinates at different resolution levels. A dissimilarity function is defined and used for matching objects to their models. Rather than considering every point on the representation, only extrema are used in this dissimilarity function. This makes the matching process efficient and less sensitive to small variations in the representation. A study of the effects of using different wavelets and their order/vanishing moments is carried out. Experimental results show that the representation is robust and, combined with the matching algorithm, it efficiently classifies unknown objects.