基于纹理合成的打印文件鉴别

为解决现有打印文件鉴别方法需要相同字符匹配的条件限制,提出了一种基于纹理合成的打印文件鉴别方法。通过对已知纹理样本块的操作,将字符图像中的空白部分修复,得到完整的纹理图像,从而消除字符内容的干扰。在打印文件鉴别实验中,通过提取完整纹理图像的灰度共生矩阵特征,并对两份打印文件的特征距离进行分类判别,最后取得了98%的正确率。

Printed Document Identification Based on Texture Synthesis

The current printed document identification methods are limited by the condition of requiring the same character to compare. In order to address this problem, a printed document identification method based on texture synthesis is proposed. The blank parts of the character image are filled through copying the sample block of the know texture. Thus a complete texture image which is not affected by the content of character is obtained. Through extracting the gray level co-occurrence matrix (GLCM) feature of the complete texture images and calculating the feature distance of two documents, the recognition rate is 98% in the printed document identification experiment.