基于多分辨几何特征的维吾尔文脱机签名识别

Research on off-line Uyghur signature recognition technology based on multiresolution geometric features

对维吾尔文手写签名图像进行二值化、去噪、归一化和细化等预处理的基础上,结合维吾尔文手写签名的结构与书写风格,对每幅签名图像进行金字塔式分辨率子图像切分,对高分辨率层抽取了共16维方向特征,对低分辨率层则抽取了共32维局部中心点特征。基于这两种特征的签名识别率分别为95.50%和90.50%。为了进一步提高识别率,又对两种特征进行了融合,结果识别率提升到了98.50%。对比分析了基于欧式距离和卡方距离度量方法对识别率的影响,确定最佳度量方法。

In this paper, on the basis of preprocessing procedures such as binarization, noise removing, normalization and thin- ning, each Uyghur handwritten signature image is segmented into several sub images with Pyramid resolution to combined with the structure and writing style of the signature, the 16-dementional directional features are extracted in higher resolution layer, while 32-dementional local central point features are extracted in the lower resolution layer. 95.5% and 90.5% of recognition rates are obtained using the two features. In order to further improve the recognition rate, the two features are combined togeth- er, and then the recognition rate is increased up to 98.5%. The effectiveness of Euclidean distance and Chi-square distance based measurement methods to the recognition rates are also analyzed, and it is confirmed that Chi-square distance is the best measure- ment method in this paper.