基于小波变换和不变矩的脱机签名鉴别

在以前研究的基础上，本文针对脱机中文签名的特点，提出利用小波变换及特征不变矩相结合的方案进行鉴别。与通常的小波变换提取细节信息不同，本文提取的是近似信息，反映了签名灰度图像灰度分布的走势。另外再结合利用伪Zemike不变矩计算的静态特征与在高密区域(HDR)基础上计算的其他动态特征。组成17维特征向量，构建脱机签名鉴别系统，并利用290个签名进行实验，实验结果表明FAR(错误接受率)和FRR(错误拒绝率)可分别达到7．83％、6．88％。

Off-line Chinese Signature Verification Based on Wavelet Transform and Invariant Moments

Based on former research, this paper proposes a new scheme for off-line handwritten Chinese signatures verification with wavelet transform and moment invariants. When wavelet transform is utilized, first grey scale image of signature is obtained through thresholding the original scanned image by using Ostu algorithm. Then weighted normalized histogram is calculated, and a 4-level dyadic discrete wavelet transform is implemented on it with Daubechies(4). 4th level approximate coefficients are reconstructed. Unlike usual application of wavelet transform to extract detail information, approximate information is obtained, which implies the grey level distribution of signature image. Then a ratio of sum of pixels with grey level lower than that of pixels with the maximum of reconstructed approximate coefficients to sum of all pixels except white ones is computed. And another ratio of distance of grey level value from the lowest grey level value to that of the highest grey level value from the lowest one is calculated. At last a ratio feature of one ratio to the other is computed. Combined with other features including static features which are pseudo-Zernike invariant moments and other dynamic features, they constitute a eigenvector of 17 features. A verification system for off-line signatures is developed, and it's tested using 290 signatures. Experiments result shows that FAR(false acceptance rate) and FRR(false rejection rate) can achieve 7.83% and 6.88% respectively.