Uighur Handwriting Identification Based On Directional Proportion Features

Handwriting identification is a technique that aims to decide the identity ofwriters according to the handwriting styles. With the improvement of applied field,handwriting identification becomes a very active area of the computer vision andpattern-recognition. At the same time, the uyghur handwriting identification is still inits infancy, the uyghur handwriting identification method research has importantsignificance, especially in ethnic minority areas, a mature uyghur handwritingidentification method is more available to people’s life.This paper mainly studies the content is as follows:(1) Collecting handwritings of the60Minority uighur students of XinjiangUniversity, devide the handwriting ogeach person into two pieces, one asthe reference sample, and the other as the test sample. then we scannedinto image and stored, formed a reference sample library and a test samplelibrary. Before reflecting features of handwriting image, first of all,wemust carry on the image preprocessing to remove a series of interferencefactors in handwriting identification. Handwriting image’s preprocessingpart mainly includes: r remove red grid of the manuscript、graying、devoicing、binarization and so on. In this part, the article lists the variouscolor model, and application of the HSV model to remove grid lines,whichmade improvements on the applicability. Corresponding feature extractionalgorithm of this paper,we also need to extract the edge of handwritingimage. This article describes severral classic handwriting edge extractionalgorithms: Roberts operator、Sobel operator、Prewitt operator、LoGoperator、Canny operator, through analyzing experimental results andcomplexity,Sobel operator was chosed in the extract of Uighurhandwriting edge. (2) Based on edge image, the author used a recursive system. Firstly, wecounted the edge pixel’s8-neighborhood pixels to get its proportion offour dimensional directions. Secondly, did this process on all the edgepixels of local windows.At last, obtained the handwriting’s proportion ofdirections which we made as feature structures.(3) Classifier processing stage, Distance function formula was used tomeasure the distance between handwriting characteristic vector. This paperlisted several common distance formula: Euclidean distance, Chi-squaredistance,Weighted Euclidean distance,Weighted Chi-square distance,According to the the characteristics of feature structures,the author alsoused the density of stroke to weight weighted Chi-square distance formulaagain to measure the similarity between handwritings. With a strongpracticability, the method achieve a better identification result.(4) We identified the degree of dependence of the sample space、therelationship between identification result and the The number ofprobability intervals, pointed out that the requirements of the objectivefactors for identification system.