Two-dimensional object recognition using a two-dimensional polar transform

摘要

A two-dimensional (2D) transform is proposed for the classification of planar objects. With a centroid referenced polar representation it samples the multiple intersections of N radii with the object. Using the mass center, translation invariance is achieved and, with an appropriate amplitude normalization, the transform is also made invariant to scaling. The modules of its 2D Fast Fourier Transform coefficients, which are rotation invariant, are used as the feature vector. Simultaneously, translation, scaling and rotation parameters are estimated. The high performance of this algorithm is shown with computer simulations, and its advantages and disadvantages are analysed in comparison with existing methods.