基于多特征协方差矩阵的图像匹配算法  

Image matching algorithm based on covariance matrix of characteristics

作　　者：王怀宝[1] 张杰[1] 张旭光[1] 韩广良[2] 王明佳[2] 董期林[3] 

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]中国科学院长春光学精密机械与物理研究所,吉林长春130022 [3]西安应用光学研究所,陕西西安710065

出　　处：《计算机工程与设计》2013年第9期3190-3194,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(61271409;61172111);中国博士后科学基金项目(2012M510768);中国留学基金项目(2011813018)

摘　　要：针对灰度及红外图像的匹配过程中经常出现的一些问题,如缺乏丰富的目标特征、易遭受复杂背景及噪声等外界因素干扰、目标出现放大缩小或偏转等,抽取目标图像的梯度幅值与方向,腐蚀与膨胀以及信息熵等特征,通过协方差矩阵将其融合在一起,构成新的特征模型。通过全图遍历求取矩阵间相似度距离的方法找到最佳匹配重心,将新方法与其它3种已有的匹配方法进行了对比说明。实验结果表明:在灰度图像匹配时新方法准确率高、鲁棒性好,同时也可以应用于红外图像中,满足了在一些条件下提高匹配准确度的要求。Aiming for the problems often appearing during the matching process of gray and infrared images., the lack of rich tar- get features, the interference problems of vulnerable to complicated background, noise and zoom-in or deflection of the target and so on, a new method is proposed. Taking the characteristics from the target images, such as magnitude and direction of the gr~ dient, corrosion and expansion of the morphology, information entropy etc. , then they are fused in together through the covaria- nce matrix, and a new feature model is gotten. Then the best points with calculating the similarity distance to traverse can be found. Finally, the proposed method is compared with the other existing matching methods. Experimental results show that the proposed algorithm keeps good performance in accuracy and robustness during the matching of gray images, and can be applied to infrared images. The requirements of the matching is satisfied, which improves accuracy in some specific conditions.

关 键 词：目标匹配 图像特征 协方差矩阵 熵 相似度距离 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术]