New Method of Feature Extraction Using Wavelet Transform and DCT in OCR
Cao Jian-hai, Changhou Lu · Journal of Optoelectronics·laser · 2004
Considering the invariant nad the noise sensitivity of feature vectors,a new method based on the wavelet transform and the discrete cosine transform(DCT) was presented to extract features in character recognition.The circular projection algorithm was used to transform a character image with two independent variables into a function of one independent variable in the circular projection space.This representation of pattern in the circular projection space will be distorted nonlinearly in the presence of noise and variety.To solve the nonlinear distortion problem,a measure distance based on the wavelet and the DCT was proposed as a nonlinear metric.The derived one-dimensional pattern was decomposed a set of wavelet transformation subpatterns with Daubechies′ wavelet transformation.The feature vectors for the original two-dimensional pattern were readily computed to use the DCT.As an application,the tag pressed and printed characters were recognized by this method.The reslts show that the proposed feature vectors can yield an excellent classification rate on the condition of noise and deformity.So,this method is proved to be in scale and orientation invariant,insensitive noise.