异源图像的多分辨率精细配准研究

发布时间：2018-12-23 19:58

【摘要】：随着成像技术的发展,各类成像设备逐渐得到了广泛应用。不同光谱或不同成像机制的成像设备拍摄的异源图像,例如多光谱卫星遥感图像、红外监控视频图像等,能够提供不同光谱的信息,同一场景下的这些不同信息通过有效“融合”能够有效提高对图像中信息的利用效率,为人类获取信息和作出决策提供重要的来源和参考价值。对图像信息的充分合理利用具有重大意义。精确的图像配准技术是获得高质量的异源融合图像的必要前提。基于局部特征的算法由于有更好的准确性和鲁棒性,是目前主流的配准方法。相对于同源图像的配准,异源图像由于局部特征会有所不同,主流的局部特征配准方法在配准精度方面会有显著下降。同时,随着图像分辨率的大幅提高以及为提高算法精度引入的大量图像后处理机制,配准算法计算量也大幅增加。有鉴于此,需要在现有技术的基础上进一步改进异源图像的配准方法以提高配准精度并降低配准过程的时间。本文通过深入研究主流图像配准的过程,分析了异源图像的局部特征差别、主流方法对于异源图像配准精度下降的原因以及配准过程的计算复杂之处。提出了基于多分辨率技术的异源图像配准方法。该方法采用全局信息作为图像配准的测度以提高配准精度,引入多分辨率机制用于剔除错误的特征匹配点对以降低计算量。通过与主流的图像配准方法比较,全局信息测度能够有效降低异源图像特征差异带来的影响而提高异源图像配准的精度,引入的多分辨率机制能够降低由此带来的额外计算量。本文提出了针对异源图像的多分辨率配准方法,并且与现有的国内外主流的配准方法进行了比较,同时将实验结果在准确性和实时性方面进行了研究、分析与总结。本文的实验结果证明,本文提出的多分辨率配准机制能够在保证与全局信息为测度的配准方法精度相当的基础上,大幅降低计算量即减少配准过程所消耗的时间。相比于主流的同源图像配准方法,本文提出的方法对异源图像的配准有较好的准确性和实时性。
[Abstract]:With the development of imaging technology, various imaging equipments have been widely used. Heterogeneous images taken by imaging devices with different spectra or different imaging mechanisms, such as multispectral satellite remote sensing images, infrared surveillance video images, etc., can provide information on different spectra. These different information in the same scene can effectively improve the efficiency of using the information in the image through effective "fusion", and provide an important source and reference value for people to obtain information and make decisions. It is of great significance to make full and reasonable use of image information. Accurate image registration is a prerequisite for obtaining high quality heterogeneous fusion images. Because of its better accuracy and robustness, the local feature based algorithm is the mainstream registration method. Compared with the registration of the homologous image, the local feature registration of the heterologous image will be different, and the registration accuracy of the mainstream local feature registration method will decrease significantly. At the same time, with the significant improvement of image resolution and the introduction of a large number of image post-processing mechanisms to improve the accuracy of the algorithm, the computation of the registration algorithm is also greatly increased. In order to improve the registration accuracy and reduce the time of registration process, it is necessary to further improve the registration method of heterogeneous images on the basis of existing techniques. By studying the process of mainstream image registration, this paper analyzes the difference of local features of heterologous image, the reason of the decline of registration accuracy and the complexity of the registration process. This paper presents a method of heterogeneous image registration based on multi-resolution technology. In this method, global information is used as the measure of image registration to improve the registration accuracy, and a multi-resolution mechanism is introduced to eliminate the wrong matching points in order to reduce the computational complexity. Compared with the mainstream image registration methods, the global information measure can effectively reduce the influence of heterogenous image feature differences and improve the accuracy of heterologous image registration. The introduction of multi-resolution mechanism can reduce the additional computation. In this paper, a multi-resolution registration method for heterogeneous images is proposed and compared with the existing mainstream registration methods at home and abroad. At the same time, the accuracy and real-time performance of the experimental results are studied, analyzed and summarized. The experimental results show that the proposed multi-resolution registration mechanism can greatly reduce the computation cost and reduce the time consumed in the registration process on the basis of ensuring the accuracy of the registration method with global information as a measure. Compared with the mainstream homologous image registration method, the method proposed in this paper has better accuracy and real-time performance for the registration of heterologous images.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】