小波变换与多特征融合的多文种离线签名识别

Multilingual off-line signature recognition based on wavelet transform and multi-feature fusion

为进一步提高多文种离线手写签名识别率,提出一种小波变换与多特征融合的离线签名识别方法。首先对采集完的签名图像进行预处理和小波变换,然后对小波变换后的4张签名子图像(低频信息图像、水平高频信息图像、垂直高频信息图像和对角高频信息图像)分别提取多尺度块局部二值模式(MB-LBP)、局部相位量化(LPQ)、韦伯描述符(WLD)和ASM能量特征,并对所提特征进行串联融合,用PCA降维形成适合的特征向量。通过训练支持向量机(SVM)和随机森林(RF)分类器对签名图像进行分类对比。文中数据集包括4个文种(英文、汉文、维吾尔文、柯尔克孜文),在单个文种识别率最高达到98%,两两混合的识别率最高达到97.54%,3个文种混合的识别率最高达到96.78%。实验结果表明,提出的方法能够有效识别多文种混合的离线签名,得到了较高的识别率。

In order to further improve the recognition rate of multilingual off-line handwritten signature,an off-line signature recognition method based on wavelet transform and multi-feature fusion is proposed.Preprocessing and wavelet transform are performed on the acquired signature images.The multi-scale block local binary pattern(MB-LBP),local phase quantization(LPQ),Weber local descriptor(WLD)and ASM(angular second moment)energy features are extracted from the four signature sub images(low-frequency information image,horizontal high-frequency information image,vertical high-frequency information image and diagonal high-frequency information image)after wavelet transform.And then,the extracted features are fused in series,and PCA(principal component analysis)is used to reduce the dimension to form the suitable feature vector.By training support vector machine(SVM)and random forest(RF)classifier,the signature images are classified and contrasted.The data set in this paper includes four languages(English,Chinese,Uighur and Kirgiz).The proposed method′s recognition rate for single language reaches 98%,its recognition rate for a mixture of any two languages reaches 97.54%,and its recognition rate for a mixture of any three languages reaches 96.78%.The experimental results show that the proposed method can effectively identify the off-line signature of multilingual mixture and get high recognition rate.