A Wavelet-based Statistical Method for Chinese Writer Identification

Writer identification is an effective solution to personal identification, which is necessary in many commercial and governmental sections of human society. In spite of continuous effort, writer identification, especially the off-line, textindependent writer identification, still remains as a challenging problem. In this paper, we propose a new method, which combines the wavelet theory and statistical model (more accurately, generalized Gaussian density (GGD) model), for off-line, text-independent writer identification. This method is based on our discovery that wavelet coefficients within each high-frequency subband of the handwritings satisfy GGD distribution. For different handwritings, the GGD parameters vary and thus can be selected as the handwriting features. Our experiments show this novel method, compared with two-dimensional Gabor model, one classic method on off- line, text-independent writer identification, not only achieves much better identification results but also greatly reduces the elapsed time on calculation.