Shape discrimination by radial sectoral projection signatures

A radial-sectoral projection (RSP) approach to shape analysis is presented. The proposed approach is RST invariant. In RSP, a shape translation technique based on the centroid is employed to center the image as in other methods. An alternative to fixed zeroeth order moment normalisation is proposed, namely an average radius normalisation. The orientation invariance is achieved by a modified distance measure. A study of the proposed method on a large character set was performed with recognition accuracies as high as 98.5%