Handwritten Numerals Recognization Based on “Effective-rows”,Feature

The concept and extracting method of the effective-rows feature, which possesses low dimension, shift-invariance and rotation-invariance under little angle is proposed on the basis of the structure of unconstraint and handwritten numerals in this paper. Its features is also constructed. According to the result of widely verifying recognition BP network, it is demonstrated that its recognitionspeed is superior to the traditional recognition methods only based on the feature of moment or wavelet and its recognition-extract is higher than some conventional methods.