摘要：The recognition of handwritten numerals hasmany important applications, such as automaticlecture of zip co des in post o.ces, and automaticlecture of numbers in checknotes. In this paper wepresent a preprocessing metho d for handwrittennumerals recognition, based on a directional twodimensional continuous wavelet transform. Thewavelet chosen is the Mexican hat. It is given aprincipal orientation by stretching one of its axes,and adding a rotation angle. The resulting trans-form has 4 parameters: scale, angle (orientation),and position (x,y) in the image. By fixing some ofits parameters we obtain wavelet descriptors thatform a feature vector for each digit image. Weuse these for the recognition of the handwrittennumerals in the Concordia University data base.We input the preprocessed samples into a multi-layer feed forward neural network, trained withbackpropagation. Our results are promising
