针对红外与可见光图像尺度差异大、图像较模糊,且图像中可提取的特征点数量不足以及特征分布不均匀易导致现有配准算法失效的情况,提出一种结合滚动引导滤波和相位信息的红外与可见光图像配准方法。利用滚动引导滤波构建尺度空间,在不增加耗时的前提下尽可能保持图像边缘信息; 提出一种改进的Shitomasi角点检测算法,提取具有尺度不变性且分布均匀的强角点; 在特征描述阶段,给出一种新的加权函数进行频率扩展,以得到更显著的相位一致性信息,实现更准确的图像特征描述。实验结果表明,该配准方法对存在9倍尺度差异的红外与可见光图像仍能实现准确配准,且对多组图像对的配准RMSE误差均保持在2像素之内。

Due to the scale as well as the resolution between infrared and visible images are quite different, the number of images feature points is insufficient and their distribution is nonuniform, which makes the existing algorithm usually fail to register the two heterotopic images, thus an improved registration algorithm is proposed by using rolling guided filter and phase information. Firstly, the rolling guidance filter is used to construct the scale space, so as to keep the image edge information as much as possible without increasing the time consumption. Then an improved Shitomasi algorithm was proposed, which extracts strong corner points with scale invariance and uniform distribution. In the stage of feature description, a new weighted function was given to spread the frequency to obtain more significant phase congruency and it’s more accurate for the feature description of blurred images. The experimental results show that the registration method can still achieve accurate registration for infrared and visible images with 9-fold scale difference, and the RMSE of the multiple image pairs remains within 2-pixel error.