摘要
针对复杂的纸币图像,采用马尔可夫(Markov)转移矩阵对图像像素间相关性进行建模,提取图像的纹理特征。鉴于各种面额的人民币图像整体纹理的空间分布比较相似,截取图像的特征区域,结合网格法,以每个网格内的纹理为特征识别纸币的面额。在纸币的训练样本中加入大量的无标注样本,用支持向量附近的无标注样本点调节分类超平面,提出改进的直推式支持向量机(Modified Transductive Support Vector Machine,MTSVM),加快了算法的收敛速度。以MTSVM为识别分类器进行仿真实验。实验结果表明在网格法的基础上,马尔可夫矩阵提取的纹理特征能够有效地描述图像,MTSVM可以得到更加准确的分类超平面,取得了较高的纸币面额识别率。
Aim to complicated banknote images, Markov transition matrix is applied to derive correlations between the consecutive pixels in digital image to extract the textural features. The feature regions of banknote images are drawn for the similarity of space distribution in diverse RMB images. Combined with grid method, the textural features of each grid are obtained to distinguish banknote images. A large number of unlabeled samples are added to the banknote＇ s training sets. The unlabeled samples which are close to support vectors are used to adjust the Support Vector Machine （SVM） hyperplane. This algorithm speeds up the convergence rate and constructs the Modified Transductive Support Vector Machine （MTSVM）. The MTSVM is presented for the recognition of banknote images. The simulation experiment results show that Markov matrix based on grid can describe banknote images effectively, and MTSVM gets higher recognition rate for its more accurate separating hyperplane.
出处
《信息技术》
2013年第3期46-50,共5页
Information Technology
基金
2010年江苏省科技支撑(工业)计划项目资助(BE2010190)
关键词
马尔可夫特征
纸币识别
图像纹理
网格特征
直推式支持向量机
Markov feature
banknote recognition
image texture
grid feature
Transductive SupportVector Machine
