Uighur Handwriting Identification Based On Statistical Feature

Handwriting is a fixed writing style of a person due to writing habits of a longtime, everyone has his own unique form of handwriting characteristics.With its higheracceptability, handwriting identification has become a very active research topic in thefields of pattern recognition.Uighur writer identification belongs to the minoritylanguage writer identification research areas, but this kind of research is notmature.Therefore,Uighur handwriting identification research has great politicalsignificance and practical significance.This thesis is to study the offline and text-independent Uighur handwritingidentification.The characters of the Uighur language are complex and ligaturesfrequently.Single feature extraction methods can not achieve better results. Based onthe structural characteristics of the Uighur language, this thesis proposes a method toextract fusion features of the Uighur handwriting.The fusion features includetexture feature basing on local binary pattern and edge direction features basing on theprobability distribution function.The main research contents are as follows:1. Pre-proeessing is the first step of handwriting identifieation.Prior to this, thepaper collects Uygur students handwritings to establish handwriting dababase and thepreprocess handwriting image.Pretreatment is removing redundant informationcontained in the image to get the binary image and lay the foundation for the next stepwork.2. The most important thing of off-line handwriting verification is the featureextraction. The cuttent feature extraction methods for Uighur handwriting only usetexture feature or structure features based on the probability distribution function.This paper proposed a new fusion characteristics which uses texture feature based onlocal binary pattern and edge direction distribution feature based on the thought ofprobability distribution to extract texture feature and structure feature of handwriting image.And the two kinds of features are fused as a newly handwriting feature in orderto express wrter’s writing style.3. This paper uses the K nearest neighbor classifier based on the distance, fordifferent characteristics using different similarity measure distance function. At thesame time, taking into account the impact of the difference content ofhandwriting,the thesis uses a weighted calculation methods in the design of theclassifier to reduce the impact effect on the identification accuracy. Finallyconsidering the different impact of edge direction distribution characteristics and localbinary pattern characteristics to the identification rate, the paper gives the twofeatures different weights to further enhance the accuracy.The experiments show that the combination of the texture feature and the edgedirection distribution feature is better than purely study the texture feature or the edgedirection distribution feature, the handwriting identification rate is higher.