The Research And Implementation Of Handwritten Numeral Recognition Method

In the field of pattern recognition, numeral recognition is an importantdirection which is being studied. Handwritten digital has its specialcharacteristics: different character fonts having few differences，the samecharacters having a variety of wording and no context linking between thedifferent numeric characters, which make off-line handwritten numeralrecognition become a most important problem and the final goal to therecognition field.This article describes the every part of the handwritten numeral recognitionsystem, including a collection of images, pretreatment, feature extraction, and soon. Wherein, preprocessing and feature extraction are the key technologies ofthe digital identification. This paper improves and testes the methods which arebeing used in the Zip code box removal, refinement of the numbers and featureextraction.Scanning the image of an envelope, postcode box is a background imagewhile Zip code box and Zip code numbers are being compared. The Zip codebox been removed, postcode box will be dealt with as a noise. This paper takeswith a method of a filter combination to remove the border, then using openingand closing operations to remove isolated points, which reduces complexity ofremoving borders and extracts the digitals.In terms of refinement, this paper uses Hilditch classic refinementalgorithm, while we use the33improved template again to detect the skeletonimage edge points after thinned to obtain a better refinement effect.In the aspect of feature extraction, this paper uses structural features whichis based on the macroscopic characteristics of the digital characters，improvingthe feature extraction which is based on the Euler number. After removing thehandwritten floral ring and calculating the Euler number, Euler number becomesconcise. The results show that the proposed algorithm is more effective.