基于边缘最优映射的红外和可见光图像自动配准算法

针对同一场景的红外和可见光图像间一致特征难以提取和匹配的难题, 提出了一种在多尺度空间中基于边缘最优映射的自动配准算法. 在由粗至细的尺度空间中, 算法分别采用仿射模型和投影模型作为参考图像和待配准图像间的空间变换模型. 在每个尺度层上, 首先基于相位一致性方法提取两幅图像的边缘结构, 并在相应的空间变换模型下将在待配准图像中提取的二值边缘映射到参考图像的边缘强度图上; 接着采用并行遗传算法寻找一组全局最优的模型参数, 使两幅图像间的结构相似度最大. 在各层的寻优结束之后, 使用Powell算法对全局寻优后的模型参数进行局部精化. 实验结果表明, 该算法能够充分利用图像间的视觉相似结构, 有效地实现红外和可见光图像的自动配准.

An Automatic Registration Algorithm of Infrared and Visible Images Based on Optimal Mapping of Edges

To aim at the difficulties in extracting and matching of corresponding features between infrared and visible images of the same scene, an automatic registration algorithm based on optimal mapping of edges in multi-scale space is proposed. In the scale space from coarse to fine, the affine model and projective model are selected as the spatial transformation models between unregistered images and references images, respectively. At each scale layer, edge structures in both images are first extracted using phase congruency method, then the extracted binary edges of the unregistered image are mapped onto the strength edge map of the reference image based on the corresponding transformation model, and then the parallel genetic algorithm is used to search for a set of global optimal model parameters, which maximizes the structure similarity measure between the two images. After optimization at each scale, the global optimal model parameters are locally refined using Powell algorithm. The experimental results show that the proposed algorithm can take advantage of visual similar structures between images and realize automatic registration of infrared and visible images efficiently.