基于多尺度结构特征和图模型的异源图像配准

项目名称： 基于多尺度结构特征和图模型的异源图像配准

项目编号： No.61303123

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 张秀伟

作者单位： 西北工业大学

项目金额： 26万元

中文摘要： 多源图像协同处理技术，充分发挥不同成像技术的优点，在探测范围、探测精度方面具有显著优势。异源图像配准为不同类型的图像提供统一的空间基准，是影响多源图像信息处理的关键。其难点在于由于不同源图像传感器成像机理不同所引起的鲁棒相似特征难以提取问题。本项目针对多源图像特征差异性问题，在现有多源图像数据库统计分析的基础上，提出异源图像多尺度局部结构特征和全局结构特征的提取与不变性描述方法；针对大量非同名点干扰的问题，提出基于概率图模型的结构特征匹配方法，并对快速算法和评价标准进行研究。该配准模型减少了对待配准图像内容的约束、增强了适应性，为异源图像配准提供一个新方法。

中文关键词： 配准；多源图像；多源图像序列；扩散张量图像；多源信息融合

英文摘要： Multi-sensor image information cooperative processing technology has many advantages over single sensor, wider observation range and more precise results. Since it can make use of the advantages of different sensors. One key problem of multi-sensoris image processing is multi-model image registration. It provides the same space standard for different sensor. Due to different image sensor has different imaging machanism, the key point of multi-model registration is how to extract robust similar features from different sensors and how to match these features robustly. In this project,we will focus on these two points. Local structure feature and graph based global feature description are combined to do multiscale struture feature extraction and description. Considering different image machnism, local feature choosing is based on staticstic analysis of different lcoal featrue on multi-sensor image database.Struture feature matching is based on probability graph model. Fast matching and registration algorithm evaluation criterion are also reaserched in this project. By this way, our approach needs less constraints on image content and possesses better applicability. It provides a new multi-model image registration method.

英文关键词： registration；multi-modal image；multi-modal image sequence；Diffusion Tensor Imaging；multi-modal information fusion