基于梯度相似性与Rényi熵图的图像配准算法

Image Registration Based on Rényi Entropic Graph Combined with Gradient Similarity

图像配准技术是图像处理与分析中的基本任务。针对图像配准对鲁棒性强、准确性高和速度快的要求,文中提出一种基于梯度相似性与Rényi熵图的图像配准算法。该算法首先提取图像特征点集,以Rényi互信息作为目标函数,然后使用特征点集的广义近邻图来估计Rényi熵与互信息,最后将特征点梯度信息融入到配准框架中。新算法结合了特征点梯度信息的鲁棒性和Rényi熵图理论的高效性。在真实遥感图像上进行的配准的实验表明,与传统方法相比,新算法在鲁棒性、速度和准确度上都达到很好的结果,是一种有效的图像配准方法。

Image registration technology is the basic task in image processing and analysis. Aiming at the requirements of good robustness, high accuracy and fast speed for image registration, propose an algorithm for image registration based on gradient similarity and Rényi en tropic graph. The algorithm extracts the feature points from images firstly, set the Rényi mutual information as the object function. Then use the generalized nearest-neighbor graph to estimate the Rényi entropy and mutual information. At last, the gradient information be tween images is integrated into the registration framework. The algorithm combined with the robustness of feature points and the high effi- ciency of using Rényi entropic graph to estimate the Rényi entropy. The experimental results show that for the real-world remote sensing images, the proposed algorithm can achieve better robustness, higher speed and better accuracy than the traditional methods. It is an effec tive image registration method.