On-line writer dependent character recognition for Khmer based on FIR system characterizing handwriting motion

Abstract

We propose an on-line writer dependent Khmer recognition method based on time-frequency characteristics of handwriting motion. The handwriting motion can be described by two features, barycenter trajectory and its velocity of pen-tip movement. Then the barycenter trajectory and its velocity of the pen-tip movement are expanded into wavelet series to extract the time-frequency characteristics of the handwriting motion. After that, The fluctuation of the wavelet coefficients can be reduced by the FIR (finite impulse response) Wiener filter. Moreover, the FIR system characterizing the time-frequency characteristics of the handwriting motion is introduced by using wavelet coefficients of the velocity and trajectory of the barycenter with fluctuation reduced as the input and output of the FIR system, respectively. The obtained impulse response of the FIR system is considered as the individual feature for a particular character. Finally, Khmer alphabets can be recognized by using the Euclidean distance between the impulse responses obtained from the reference alphabets and those of the alphabets to be recognized. Khmer character recognition experiments were performed on a database consisting of 6,750 of numerals and alphabets written by 17 people. As the results, the average of the recognition rate was 98.17%.