基于视觉注意机制的各向异性尺度空间下多源遥感图像配准方法研究

多源遥感图像可获取观察物更丰富的信息，克服单一类型遥感图像信息不全的问题。但异源遥感图像在时间、空间、分辨率等方面存在较大差异，为整合异源遥感信息需对多源遥感图像进行配准。本项目以提高多源遥感图像配准精度为主要研究目标，以多源遥感图像为研究对象。依据生物视觉注意机制理论，构造自底向上—自顶向下的双向视觉显著模型获得全局显著图；依据物理学的热传导扩散原理，借助非线性滤波技术，构建各向异性尺度空间模型提取特征角点；构造高斯离散正交矩表征图像角点。研究和解决自适应非线性的局部显著图融合策略、扩散权重的调节策略、复合扩散传导方程模型、Gaussian-Krawtchouk不变矩的离散形式等关键问题。为得到高精度多源遥感图像配准提供有效的实验方案。本项目提出一套基于视觉注意机制的各向异性尺度空间下多源遥感图像配准建模理论和方法，可丰富图像特征配准理论，对多源遥感图像配准关键技术突破有着积极意义。

Multi-source remote sensing images include more comprehensive information about observations. They can overcome the shortcomings of incomplete information from a single remote sensing image. They have broad application prospects in the field of remote sensing. Due to the great differences in time, space and resolution between multi-source remote sensing images, the stage of image registration is indispensable in order to integrate the heterogeneous remote sensing information. Improving the registration accuracy is taken as the main research target in our project. The multi-source remote sensing images are taken as our research objects. According to the theory of biological visual attention mechanism, a bottom up—top down bidirectional visual attention model is constructed to obtain global saliency images. Based on the physics principle of the heat conduction diffusion and with the help of the nonlinear anisotropic filtering technology, we build a mode called pyramid of anisotropic angular scale space which is used to extract features. And then, the new Gaussian discrete orthogonal moments are constructed to describe the character of image. .We deal with the key problems such as a strategy of adaptive nonlinear local saliency graph fusion, a strategy of the diffusion weight adjustment, the mode of compound diffusion conduction and deduction of discrete from derivation of Gaussian-krawtchouk invariant moments. We provide an effective experimental scheme for the higher precision in multi-source sensing image registration. The project proposes the theory for multi-source sensing image registration based on the visual attention mechanism under the background of the anisotropic scale space, which can enrich the image registration theory. And it is also positive significance to the core technology breakthrough for high precision and efficient multi-source remote sensing image registration.