In this paper new solution of handwritten digits recognition system is presented. System is based on CDWT - Complex Discrete Wavelet Transform and MP - Matching Pursuit decomposition. CDWT decomposition generates coefficients for base of feature vectors. Instead of standard median filtering in pre-processing stage we propose Matching Pursuit Decomposition. MP is based on BFD - Base Functions Dictionary which consist of elementary 2D functions (atoms). The proposed BFD has good properties in case of noise reduction. Method presented in article utilize dual tree CDWT decomposition in a different way than in patent . Patent  uses dual tree CDWT transformation in more general way in order to identify features in dataset such as images and one dimensional signals (phase difference is indicative of a different type of feature).
Handwritten digits recognition, CDWT, matching pursuit, base functions dictionary, Matching Pursuit decomposition, Discrete Wavelet Transformation, Complex Discrete Wavelet Transform, Pyramidal Gabor Wavelets, Magarey Wavelets, 2D spectral analysis, Robust motion estimation, Pattern Recognition, Image Processing, complex wavelets
Institute of Telecommunications, University of Technology&Life Sciences, Kaliskiego 7, 85-796 Bydgoszcz, Poland