Research Of Image Registration Method Based On Bayesian Model

In recent years,with the rise of smart wearable devices,such as VR,AR and other related equipment,and promote the development of image processing and computer vision and other related technologies,image registration technology has also developed.And image registration is very extensive,such as militar y,aviation,medical,robot and other fields,image registration technology plays a very important role.In this paper,we propose a method combining Bayesian theory with Markov chain Monte Carlo algorithm to realize image registration.According to Bayesi an theory,the corresponding image registration model is established,and then the Markov chain Monte Carlo algorithm is used to solve the parameters of the image registration model.In the establishment of image registration model,by introducing the conc ept of "accurate image",the "bridge" between the images to be registered is built,and the information parameters of image registration are successfully introduced into the correlation function,including spatial transformation Information,and the corresponding parameters between the images,and then according to Bayesian theory,the establishment of the relevant likelihood function and a priori distribution function,and finally get the image registration model.In the solution of the space transformation parameters,the posterior probability function of each parameter is obtained by making the relevant agreement on the prior information of some parameters.In this paper,we use the theory of von Mises distribution and conjugate distribution,Function of the parameters,the use of Markov chain Monte Carlo method for iterative update.Finally,in the search for the relationship between the corresponding points,this paper uses a random walk method to simulate the distribution of the corresponding points,through the Metropolis-Hastings method to derive the probability of acceptance in each case,so as to accelerate the algorithm to update the most iterative The distribution of the corresponding points.The experimental results show that the proposed algorithm can achieve image registration and can meet the requirements of image registration.