Image registration with position and similarity constraints based on genetic algorithm

This paper presents an original usage of genetic algorithm in application to 2-D aerial image registration with position and similarity constraints. The framework of the presented method includes four steps: feature points extraction, mesh generation, mesh registration and image registration. First, the position and inner position constraints are selected based on the visual features manually or automatically on the aerial image and the corresponding satellite or map image. Second, the two images are triangulated with position constraints. Third, Genetic algorithm is applied to optimize the similarity measure based on the absolute orientation technique together with the position constraints. The object function is based on the similarity measure on corresponding pairs of triangles in the meshes. Finally, the registered meshes are used to obtain the image registration. Experimental results on simulated data and interesting applications have demonstrated that the proposed algorithm can generate accurate results efficiently.