基于非参数搜索的打印墨粉纹理合成及评价

Synthesis and Evaluation of Toner's Texture Based on Nonparametric Search

打印墨粉纹理合成是基于纹理合成的打印文件检验技术的基础步骤;而定量地评价纹理合成的效果是提高打印文件检验准确率的关键。采用基于马尔科夫随机场的非参数搜索算法将打印墨粉纹理子图合成为完整的纹理图像,将其作为训练样本。将每幅训练样本分割为16幅子图像,分别再次合成作为测试样本。对训练样本和测试样本提取灰度共生矩阵特征,接着利用主成分分析将样本特征维数进行压缩,得到最具鉴别力的特征分量。最后,计算训练样本和测试样本的加权欧氏距离,进行图像匹配,得到测试样本的相似率,从而定量地评价了纹理合成算法的效果。对4种打印墨粉纹理子图进行实验,测试样本的相似率均在95%以上,证明了非参数搜索算法的有效性。

Synthesis of toner＇ s texture is the fundamental step of print document examination technology based on texture synthesis. Quantitative evaluation of the effect of the texture synthesis is the key to improve the accuracy of print document examination. Nonparametric search algorithm based on Markov Random Field is used to synthe- size the subgraph of toner＇ s texture into a complete texture image as the training sample. Each training sample is divided into 16 sub image and each sub image is synthesized as the test sample. Gray level co-occurrence matrix feature of training samples and test samples are extracted. Then, principal component analysis is used to compress the dimension of samples＇ features to obtain the most discerning features. At last, image matching is conducted by calculating the weighted Euclidean distance between training samples and test samples to get the similar rate of test samples, which evaluates the effect of the algorithm of texture synthesis quantitatively. Experiments are conducted on four subgraph of toner＇ s texture and similar rates are all above 95 %, which proves the validity of the nonpara- metric search algorithm.