CNN Templates Design Based On The Optimized Particle Swarm Algorithm And Application In Image Registration

With the progress of science and technology and the arrival of the information age, the information’s processing speed problem has become one of the urgent key issues. People have to deal with the larger quantity of information, the real-time requirements are also more and more strongly, speed and real-time processing in the information processing information has been two big difficult problems to solve urgently for the person, especially in the image processing, the images contain a large quantity of information, which make many mature algorithms get to the bottleneck in the application of the process of image processing due to speed questions.As we all know, the parallel operation is the most effective way to solve the speed problem, so finding the parallel processing function algorithm into the researchers’ sighting.The CNN is very suitable for information processing and calculation with parallel structure processing; it has been widely applied in the image processing, video signal processing and so on. So the research and design of the CNN template has very important theory meaning and practical value. In this paper the design method is improved for the CNN template and its application in medical image registration with image processing. The main innovation of this paper is:(1) Designing CNN template based on the particle swarm optimization. The parameter value selects will appear local extreme value instead of the global optimal value problem in design process the CNN template. By optimizing the parameters of particle swarm optimization algorithm, the CNN template value is the optimal parameter. It won’t appear the characteristic value extraction errors occurred in process of image feature extraction, making the image registration process is more accurate.(2) Image registration algorithm based on mutual information of the CNN. About the having image registration algorithm, it included the characteristics and mutual information.They have advantages and disadvantages for example the high computational complexity and the amount of information inaccuracy etc in the registration. In this paper, the two kinds of algorithm are combined. It includes the analysis of image features with basing on mutual information image registration algorithm. It is placed in the image registration process and makes the calculation in the process of registration of the complexity is reduced. At the sametime the information gain and improve the certain supplement, making the registration results in accuracy better and speed faster.(3) Image registration algorithm based on mutual information and the CNN of parameter optimization. Above proposed CNN mutual information image registration algorithm and the two kinds of algorithms of image registration are combined. But it still exist some problems not to solve. The use of CNN for feature extraction will fall into extreme selection difficult maximum mutual information, and the need in the process of registration, relates to the parameters in the process of mutual information value choice. Face the problem of the above problems, optimum calculation parameters by using the particle swarm algorithm separately on the CNN template parameter and the mutual information values involved, further reducing the computation complexity at the same time. Solves the problems of CNN mutual information image registration algorithm in the problem, so that the accuracy of image registration results higher efficiency and faster.