Chinese handwriting-based writer identiication with PDTDFB transform

Abstract

In order to enhance the accuracy of Chinese off-line handwriting recognition, a new method based on the pyramidal dual-tree directional filter bank (PDTDFB) was presented. According to multi-resolution, arbitrarily high direction resolution, low redundant ratio and efficient implementation properties, the PDTDFB transform can effectively capture more edges and contours in image. Using the extracting features with GDD model to measure the KL distance, we get the image retrieval precision rate. In comparison to the scalar wavelet transform, the complex wavelet transform (CWT) and Contourlet transform, the method increases the accuracy about 22.3%, 7.5%, 2.3%, separately.