Phase referenced image dictionary for handwritten character recognition

This paper proposes a novel scheme for handwritten Odia character recognition based on the phase information. For the purpose, an iterative phase reconstruction based recognition scheme for isolated Odia handwritten characters has been proposed. The phase information are extracted from a set of character images belonging to a particular class using the fast fourier transformation (FFT). Suitable number of iterative reconstructions have been applied on the images. A phase referenced image (PRI) is created out of these set of reconstructed matrices. The same procedure is applied to all the fifty-seven classes of characters (alphabets and numerals) and their phase referenced images are stored into a PRI dictionary. For testing the class label of a probe handwritten character, the cross-correlation of it's corresponding reconstructed phase matrix is found out against the PRI dictionary. The least value of this cross-correlation is used to determine the class label of the probe character. To validate the efficiency of the proposed scheme, it has been compared with three other competent existing schemes in terms of accuracy and computation time. The proposed scheme shows an overall rate of accuracy of 87% with a reduced computation overhead.