Research On Image Registration Technology Based On Mutual Information

Image registration is the basis and prerequisite for realizing image fusion. It has been widely used in remote sensing image processing, automatic navigation, medical diagnosis, computer vision and pattern recognition, optical and radar image target tracking and other fields. How to make optimization strategy more efficient, similarity measure more robust, how to avoid local extremum, how to improve the image registration precision and speed, to get high quality images and meet the massive data's real-time processing demand, becomes a research focus area for image registration in present stage. In order to improve precision and speed up the convergence of image registration, we focus on image registration optimization algorithm based on the normalized mutual information, also summarize the advantages and disadvantages of classic algorithm, then propose an improved algorithm, and simulation experiment. The main work is as follows:Firstly, the concept of image registration, current domestic and international research background, present situation and significance are comprehensively narrated, then the principle of image registration, basic model, application classification and image registration method are described systematacially;Secondly, the key problem of image registration is similarity measure, which is used to measure the degree of alignment of two images. Optimize search strategy directly relate to the accuracy and efficiency of image registration. We mainly introduce the mutual information measure function and optimization algorithm. Considering NMI is susceptible to the affect of noise, we propose a kind of normalized mutual information adaptive combined with image's edge, experiment proves its feasibility. By comparing some classic optimization algorithm, the experimental result shows the importance of optimization algorithm;Thirdly, using NMI as similarity measure, we propose a new hybrid search optimization strategy to solve the transformation parameters for registration. The new algorithm uses the idea of PSO algorithm, to improve the pheromone updated rules of continuously ant colony algorithm, combines with Powell algorithm. Experiments show that the hybrid algorithm can ensure the global convergence, has high registration precision, fast registration speed, and it has the better usability.Forthly, an adaptive accelerate particle swarm optimization algorithm based on the accelerated factor is proposed, and applies to optimize image registration based on the NMI. By sorting the solvings, a specified number of worst solutions will be forced to accelerate to the direction of the global solution, and improve adaptive inertia weight formula, thereby it improves the convergence, prevents premature convergence and increases the diversity of the optimal solution, while algorithm adds the accelerated factor to improve convergence speed. Experiments get a satisfied registration results.