The Study Of Image Matching In Navigation System Based On Mutual Information And Particle Swarm Optimization Algorithm

Firstly, the basic principle and steps of image matching are briefly introduced in this paper. The image matching methods based on gray are compared with the methods based on feature by analyzing their respective characteristics. In this foundation, mutual information algorithm is studied with its principle, the related concepts, the application in image matching and the process of its realization. The main purpose of this research is to study the adaptability of mutual information similarity measurement in remote sensing image matching.Secondly, in order to improve the correction of image matching, the effect of different gray levels on mutual information is discussed later. At the same time, some indexes which used to describe the quality of image such as SNR, gradient and repeated pattern are used to analyze the relationship between mutual information and the quality of image. The result shows that mutual information can get better results than normalized cross-correlation coefficient because it ignorance the influence of gray no liner distortion.Thirdly, because of the low efficient of mutual information algorithm, a new global optimal method called particle swarm optimization is used in the process of image matching which applies to mutual information algorithm. It's more difficult for navigation image matching that the optimal method will be easily fall into local-maxima, while mutual information algorithm may also cause many more local-maxima in different sensors and different resolutions image matching. So, an improved method of the primal particle swarm optimization has to be used to solve the problem referred in this article.The optimal parameter setting, end of loop condition and the enhancement of some improved methods on particle swarm optimization algorithm are discussed in this paper. During the experiment, it's found that the methods based on stochastic initialization, constrained coefficient and distribution with uniform designing can improve the correction of image matching largely without consuming too much calculating time. The study of manners and mechanism of stochastic initialization is discussed and a new integrated particle swarm optimization algorithm based on stochastic initialization is put forward later, which is preferred to get better results in multi-source remote sensing image matching.The integrated method of mutual information and improved particle swarm optimization algorithm is approved to be applicable in the area of remote sensing image matching while using multi-resource remote sensing images from different sensors as experiment images. Compared to the traditional method of normalized cross-correlation coefficient, mutual information algorithm performs to be more robust and accuracy, while combined with the integrated method of particle swarm optimization algorithm based on stochastic initialization to enhance the matching efficiency.