Research On Uyghur Handwirting Identification Technology Based On Structural Features

Writer identification is the task of determining the author of a sample handwriting froma set of writers. It has widespread application domain of insurance, finance, public security,archeology and so on, because of this technique has many advantages such as free ofinspectors’ subjective factor, high identification rate and quick distinguishing etc. Manyreports about writer identifications are mostly based on Latin handwriting, Chinesehandwriting and Arabic handwriting. However, there have been some reports for automaticwriter identification based on Uyghur handwriting, but the identification result is not ideal,and it is still need a lot of research for practical use. Therefore there is still much room forresearch on Uyghur handwriting based writer identification, because of the technique ofethnic minority’s automatic writer identification in our country is still in the state ofpreliminary stage.In this paper, the exploratory study for the offline Uyghur text-independent writeridentification technology is conducted based on the structural features of the Uyghur andhandwriting features. The focus of this study is the feature extraction of Uyghur handwriting.The main aim of the research is including to identify the characteristics and extractionalgorithms suitable for Uyghur handwriting, its effectiveness and stability analysis, and todevelop the experimental platform of Uyghur writer identification.The Uyghur handwriting sample database which include1344handwriting sampleswritten from224individuals are constructed firstly. Then carried out the preprocessingoperations such as the removing of the grid lines and noise, grayscale, binarization, row andconjoined section segmentation and thinning. According to the structural characteristics ofUyghur, whole page layout features, directional features and stroke statistical features, aswell as their extraction algorithm has also been proposed in this paper. In addition to theabove features, inclination and curvature features, which are have higher identification ratein English and Arabic handwriting, are also applied in Uyghur handwriting identification.Finally, these five types of features are classified by using distance measurement methods,such as Euclidean distance, similarity and Chi-square distance, and three different kinds ofexperiments conducted to determine the best features and distance measurement method.The experimental results show that the stroke statistical features which extracted from thethinned Uyghur handwriting obtain the highest identification rate of98.66%, and chi-squaredistance has the best classification results in this paper.