基于LBP和纹理信息挖掘的打印文件机源认定

Machine source identification of printed document based on LBP and texture information mining

为了消除光照对打印文件图像采集的影响,采用均匀模式的LBP对采集到的字符图像进行处理,提取LBP特征值,使用SVM进行打印文件的分类鉴别,分类准确率只有83.40%,说明LBP特征可以用于打印文件的分类鉴别,但是分类准确率不能满足实际应用的需要。运用两因子方差分析模型得出字符因子和打印机因子对鉴别率的影响都是显著的。采用纹理信息挖掘的方法消除字符因子的影响,研究打印机因子对打印文件鉴别的影响,可以将打印文件鉴别率提高到94.50%,具备实际应用的能力。

In order to eliminate the influence of illumination to printed document image acquisition, the uniform LBP is used to process the collected character-image and extract LBP feature values of it. SVM is applied to classify and identify printed document. The accuracy rate is only 83.40%, which indicates that LBP feature can be used to printed documents identification, but the accuracy rate is not meet the needs of practical application. According to the two-factor variance analysis model, it is concluded that the influence of the character-factor and the printer-factor on the discrimination rate are significant. Texture information mining method is used to eliminate the influence of the character-factor and the focus is put on studying the impact of printer-factor on the printed file identification. The printed file identification rate can be increased to 94.50%, with actual application ability.