ANALYSIS OF PERSONAL INFORMATION IN HANDWRITTEN CHARACTERS, AND AUTOMATIC WRITER RECOGNITION USING A SPECTRAL RESOLUTION TECHNIQUE.

This paper proposes application of a new spectral resolution technique of 2nd order statistics for the purpose of extracting personal information from handwritten characters. This resolution technique was formulated based on a textural analysis viewpoint, and put to use for automatic writer recognition. Three algorithms were developed for text-independent identification, text-dependent identification and text-dependent verification. Experiments with five characters regarding text-dependent writer identification show that the identification rate is 99. 979% for 30 persons. Moreover, in verification experiments, false acceptance was only 0. 0034% when false rejection was 5. 2% for 29 registrants and five characters.