基于纹理特征的维吾尔文离线手写签名鉴别  

Uyghur offline handwritten signature verification based on texture features

作　　者：张淑婧 麦合甫热提 吾尔尼沙·买买提 朱亚俐[1] 库尔班·吾布力[1] ZHANG Shu-jing;Mahpira;Hornisa·Mamat;ZHU Ya-li;Kurban·Ubul(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China;Educational Administration Department,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046 [2]新疆大学教务处,新疆乌鲁木齐830046

出　　处：《计算机工程与设计》2020年第3期770-776,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61862061、61563052、61163028);新疆大学2018年度博士启动基金项目(62008040)。

摘　　要：为进一步提高维吾尔文离线手写签名鉴别正确率,提出基于纹理特征融合的离线签名鉴别方法。对经过预处理的签名图像分别提取多尺度块局部二值模式(MB-LBP)和局部相位量化(LPQ)两种纹理特征,将两种特征进行串联融合,形成高维纹理特征向量。通过训练随机森林(RF)对签名图像进行分类鉴别。在共包含1800个签名图像的维吾尔文数据库和包含2640个签名图像的CEDAR数据库中得到的总正确率分别为96.35%和96.73%,结果表明该方法有效提高了维吾尔文离线手写签名鉴别正确率。To further improve the accuracy of Uyghur offline handwritten signature verification,an offline signature verification method based on texture feature fusion was proposed.Multi-scale block local binary pattern(MB-LBP)and local phase quantization(LPQ)were used to extract the features of pre-processed signature images.The features were fused in series to form a high dimensional texture feature vector.The signature image features were classified and verified through training random forest(RF).96.35%and 96.73%of overall right rate(ORR)are obtained respectively on Uyghur signature database containing a total of 1800 samples and CEDAR Latin signature database containing 2640 samples.The results show that the proposed method effectively improves the ORR of Uyghur offline handwritten signature verification.

关 键 词：维吾尔文 手写签名鉴别 多尺度块局部二值模式 局部相位量化 随机森林 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]