Directional Continuous Wavelet Transform Applied to Handwritten Numerals Recognition Using Neural Networks

The recognition of handwritten numerals has many important applications, such as automatic lecture of zip codes in post offices, and automatic lecture of numbers in checknotes. In this paper we present a preprocessing method for handwritten numerals recognition, based on a directional two dimensional continuous wavelet transform. The wavelet chosen is the Mexican hat. It is given a principal orientation by stretching one of its axes, and adding a rotation angle. The resulting transform has 4 parameters: scale, angle (orientation), and position (x,y) in the image. By fixing some of its parameters we obtain wavelet descriptors that form a feature vector for each digit image. We use these for the recognition of the handwritten numerals in the Concordia University data base. We input the preprocessed samples into a multilayer feed forward neural network, trained with backpropagation. Our results are promising.

[1]  Ezequiel López-Rubio,et al.  A principal components analysis self-organizing map , 2004, Neural Networks.

[2]  Spyros Liapis,et al.  Color and texture image retrieval using chromaticity histograms and wavelet frames , 2004, IEEE Transactions on Multimedia.

[3]  Romain Murenzi,et al.  Pose estimation of SAR imagery using the two dimensional continuous wavelet transform , 2003, Pattern Recognit. Lett..

[4]  Andrew F. Laine,et al.  Wavelet descriptors for multiresolution recognition of handprinted characters , 1995, Pattern Recognit..

[5]  Ingrid Daubechies,et al.  Ten Lectures on Wavelets , 1992 .

[6]  Anders Krogh,et al.  Introduction to the theory of neural computation , 1994, The advanced book program.

[7]  S. Mallat.  A wavelet tour of signal processing , 1998 .

[8]  Ana M. C. Ruedin.  A nonseparable multiwavelet for edge detection , 2003, SPIE Optics + Photonics.

[9]  Enrique Carlos Segura,et al.  Un clasificador neuronal que explica sus respuestas: aplicación al reconocimiento de dígitos manuscritos , 2003 .

[10]  D. Gorgevik,et al.  An efficient three-stage classifier for handwritten digit recognition , 2004, ICPR 2004.

[11]  Adam Krzyzak,et al.  Contour-based handwritten numeral recognition using multiwavelets and neural networks , 2003, Pattern Recognit..

[12]  Touradj Ebrahimi,et al.  JPEG2000: The upcoming still image compression standard , 2001, Pattern Recognit. Lett..

[13]  Jean-Pierre Antoine,et al.  Two-dimensional directional wavelets and the scale-angle representation , 1996, Signal Process..

[14]  Jean-Pierre Antoine,et al.  Target detection and recognition using two-dimensional isotropic and anisotropic wavelets , 1995, Defense, Security, and Sensing.

[15]  Ana M. C. Ruedin,et al.  Reconocimiento de dígitos manuscritos usando la transformada wavelet continua en 2 dimensiones y redes neuronales , 2006 .

[16]  Hong Yan,et al.  A nonlinear neural network model of mixture of local principal component analysis: application to handwritten digits recognition , 2001, Pattern Recognit..

[17]  Simon Haykin,et al.  Neural Networks: A Comprehensive Foundation , 1998 .