针对异源图像提出一种基于多尺度密集结构特征的快速匹配算法。算法首先利用 Gabor滤波器逐像素提取图像中的结构响应，再根据主方向响应对多尺度结构特征融合，然后使用快速傅里叶 变换在频域计算各特征分量图像之间的卷积，最后将卷积生成的系数矩阵求和计算出图像之间的相似性并选择相似性最大位置作为匹配结果输出。本文算法能有效适应异源图像间的非线性灰度变化 和噪声干扰问题。测试使用可见光、红外、雷达图像组成的异源图像数据集对本文算法和现有算法进行测试比较，结果表明：本文算法的平均误匹配率最低，并且计算速度有明显优势。

A fast image matching algorithm based on a multiscale dense structure feature has been proposed for matching multi-sensor images. In this method, the Gabor filter is employed for generating the structure response of the image. Then, the multiscale structure feature is combined on the basis of the major orientation response. Subsequently, fast Fourier transform is employed to calculate the convolution for each feature component image in the frequency domain. Finally, the similarity between images is estimated based on the sum of the convolutions, and the position with maximum similarity is outputted as the matching result. The proposed algorithm can effectively adapt to non-linear intensity variation between a multi-sensor image and noise distortion. In the experiments, a dataset consisting of optical, infrared, and synthetic aperture radar images was used for evaluating the proposed algorithm and other existing algorithms. The results indicate that the average error matching rate of the proposed algorithm is the lowest among the investigated algorithms and it has a distinct advantage in terms of computational performance.