【作者】 刘红毅；

【导师】 张书玲； 王蕴红；

【作者基本信息】 西北大学 ， 计算数学， 2003， 硕士

【摘要】 信息融合是对多源数据进行综合处理，从而产生新的、更有意义的信息。目前，信息融合技术已成为一个十分活跃的热门研究领域，并被广泛的应用于各个领域。本文研究信息融合技术在遥感图像和多生物特征身份鉴别系统中的应用。 不同传感器所获得的遥感图像包含的信息不同，为了得到更加全面、可靠的信息，需要对多传感器图像进行综合处理(融合)。本文研究的一个重点是像素级和特征级的图像融合算法。 在遥感图像的像素级融合算法中，我们研究了基于小波变换的图像融合。在对小波变换的基本理论和其用于融合的原理进行介绍之后，我们利用了灰度值选择算子针对Radarsat和Landsat图像进行了相关实验。在特征级的图像融合中，我们针对Radarsat和Landsat图像的特点，采用滑动窗方法提取适当的特征，然后用Bayes理论来进行特征层的融合分类。 近年来，生物特征的身份识别技术有了飞速的发展，基于多生物特征融合的身份识别技术也得到了更多的关注，多生物特征的有效结合提高了身份鉴别系统的性能。这是本文研究的另一重点。 在分析了多生物特征融合系统的研究现状的基础上，我们研究了有参数和无参数两类融合体系。在有参数的融合方法中，分析了Bayes理论和Neyman-Pearson准则在融合时的适用范围；针对Bayes融合系统参数的选取问题，我们分别进行了全局参数和局部参数的融合实验；并提出了加权思想，将其用于Bayes理论和Neyman-Pearson准则的身份识别融合系统，得到了较好的鉴别效果。 在研究无参数的融合方法时，我们分别进行了基于K-NN和ENN的多生物特征身份鉴别的实验，进而提出了改进的ENN算法。实验表明，改进ENN的身份鉴别融合系统的认证率比K-NN和传统ENN融合系统有所提高。更多还原

【Abstract】 Information fusion deals with multisource data in order to get more new, significant information. Today, information fusion is becoming one of the most active researches, and is widely applied to all kinds of fields. The intention of the paper is to investigate the applications of information fusion in remote sensing and multi-modal biometric identity verification system.Different remote sensing images have different information, and the combination of multisensor images (fusion) can get more complete and reliable information. One of the keystones of the paper is image fusion algorithm based on pixel level and feature level.In the pixel-based fusion algorithm, we focus on image fusion based on wavelet transform. After the introduction of the wavelet theory and the principle of wavelet-based image fusion, we test the Radarsat and Landsat images using gray-value algorithm. When it comes to feature-based image fusion, we make use of Mean-Shift to extract appropriate features according to the characteristics of Radarsat and Landsat images, then apply the Bayes theory to feature level fusion classification.In recent years, biometric identity verification has a rapid development, multi-modal identity verification has gained more and more attention, and the combination of multimodal can improve the performance of the identity verification system. The other focus is multi-modal identity verification systems.Based on analysis of the recent research on the multi-modal fusion system, we investigate the parameter and non-parameter fusion systems. In the parameter-basedfusion methods, we analyze the applicability of the Bayes theory and Neyman-Pearson rule when they are applied to identity verification systems; we compare global and local parameter in Bayes fusion system; we further propose weighted method, and apply it to Bayes- and Neyman-Pearson- based identity verification system, and get a more higher verification result.In the non-parameters fusion methods, we use K-NN and ENN classifiers to combine different biometrics, and we propose improved ENN method. Compared with K-NN and ENN identity verification systems, the performance of verification system using improved ENN is enhanced.更多还原