Automatic classification of form features based on neural networks and fourier transform

This paper focuses on the identification and classification of forms in image document management system. It introduces a methodology that uses the pretreated horizontal and vertical projection of the forms for Fourier transform and the resulted power spectrum density as the eigenvector. Then we study and practice to extract the characteristics of the forms using BP neural network. This method has overcome the deficiencies caused by poor generalization or being unable to identify symmetric form structure correctly. Experiments have proved that this method can perform classification on forms with different structures, and has excellent adaptability.