Research On Image Registration Based On Fractional Order Derivate And Feature With Inertial Constraint

Non-rigid image registration is one of the key techniques in computer vision.It is widely used in medical image processing,pattern recognition and many other applications.Traditional Demons algorithm only relies on the gradient to drive the moving image diffuse into the fixed image,and there is only intensity similarity criteria in the similarity measure function.These factors make Demons difficult to finish the registration in the weak texture or smooth region,and often lead to low accuracy.In order to solve these problems,this paper proposes a Log Demons algorithm based on the combination of side window fractional order derivate and feature with inertial constraint.We design a new side window fractional order derivate mask to calculate the image gradient to improve the gradient in the weak texture and smooth region.A similarity constraint based on Gabor feature is added to the similarity measure function of Log Demons.When calculating the updated field at the current iteration,we introduce the previous updated fields as the inertial constraint information to strengthen the current updated field.This can make up for the insufficiency of the driving force.Through the above improvements,a high accurate image registration algorithm based on Log Demons is proposed.The main work of this paper includes:1.A LogDemons algorithm based on side window fractional order derivate is proposed.According to the definition of G-L fractional order derivate,we design a new fractional order derivate mask to convolve with the image to obtain the gradient,which replaces the gradient calculated by integer order derivate used by Log Demons algorithm during the process of minimizing the similarity measure function.It can enhance the registration accuracy in the weak texture and smooth region.The way of calculating the gradient in the edge region is improved by the idea of side window filtering.The improved mask can enhance the gradient of pixels on the edge which can better drive the moving image to the position corresponding to the fixed image.2.A Log Demons algorithm based on Gabor feature similarity constraint is proposed.Gabor feature can represent the image geometric characteristics,and is robust to illumination change and noise.Log Demons is only consisted of intensity similarity criteria,and it does not take the relationship between image features into consideration which can have a positive effect on registration.The new function introduces the Gabor feature similarity and can significantly improve the geometry similarity between images.3.A Log Demons algorithm based on the inertial consistency constraint of the deformation is proposed.It is difficult for Log Demons algorithm to obtain precise updated field in regions where image gray scale is uniform.When calculating the updated field at the current iteration,we propose to integrate the previous updated fields adjacent to the current iteration into Log Demons as the inertial constraint information.The addition of the inertial constraint can effectively strengthen the updated field and reduce the influence of insufficiency of the driving force caused by the small variation of the gray scale.The method proposed in this paper is tested on three types of image sets including synthetic images,real scene images and MRI brain images.And the proposed method obtains satisfactory registration accuracy.The experiment results demonstrate that the proposed method is superior to the compared methods,such as Log Demons,LDDMM and Spectral Log Demons,which verifies that the proposed method is effective.