Research On Non-rigid Image Registration Algorithm Based On Demons

In the filed of computer vision,image registration technology is an important research direction,which is widely used in medical,astronomy,military and other fields.Image registration technology always is mainly divided into rigid image registration and non-rigid image registration according to the deformation of the image.Among them,the method based on rigid image registration mainly solves the problem of linear deformation of images.At present,the related research on rigid image registration basically tends to be mature.Then,non-rigid image registration mainly solves the non-linear deformation.the research on non-rigid image registration needs to be further developed.Nowadays it has become one of the research hotspots in image processing,so non-rigid image registration has important research value and application significance.Demons algorithm is an important non-rigid image registration method,which has received extensive attention due to its advantages of fast convergence speed and convenience.However,many scholars have improved it because it can not give consideration to image registration with small and large deformation,and needs to continuously adjust parameters according to the size of deformation.in this paper,the original Demons algorithm,Active Demons,Diffeomorphic log-demons algorithm and Symmetric Demons algorithm are deeply studied.Based on the theory of normalized mutual information image registration,The paper improves Demons algorithm and proposes Demons non-rigid image registration algorithm based on mutual information improvement.The algorithm improves the driving force of Demons algorithm by introducing normalized mutual information,which can not only give consideration to image registration with small and large deformation images at the same time,but also adaptively adjust Demons algorithm by adopting normalized mutual information of the moving image and static image,thus avoiding the complexity of parameter adjustment of Demons algorithm.The experimental results of medical images,natural images,synthetic images and binary images show that the method can better process large deformation images and preserve their edge features,thus obtaining better registration accuracy and visual effect.