Research On Non-rigid Medical Image Registration Lgorithms Based On Optical Flow Model

Medical image registration is to make characteristic points, face or pixels of twomedical images achieve space consistent through looking for some space change whichhas a very important research and clinical application significance. It is valuable.Medical image registration algorithms can be classified into rigid medical imageregistration and non-rigid registration algorithms. They were studied systematically inthis paper. In order to achieve non-rigid medical images registration, the steps andtechnologies of medical images registration are explored and practiced. Somedisadvantages of the traditional algorithms are improved. A new non-rigid registrationalgorithm based on optical flow model is proposed successfully which is morereasonable and improved.The main work in this paper as follows:(1) Based on the characteristics of non-rigid heart image deformation, a step-by-step image registration strategy is proposed. Firstly, SIFT feature algorithm isused to image to get rough registration and then optical flow algorithm is usedto achieve the final accurate registration.(2) Local image registration were achieved by using SIFT characteristicalgorithm which can overcome the changes of image scaling, rotating, scalebrightness, etc and has high recognition and strong adaptability and be benefitto the follow-up matching. SIFT features algorithm can be used in timesequence images which can extract image angular point automatically, andfast.(3) Demons registration algorithm based on optical flow model was used in thispaper and in the basis of original image deformation force, add a force whichis gradients of mutual information of two images on current migrationtransform. To make floating image transformation towards the direction ofthe mutual information increases of the two images. When mutualinformation of the two images no longer increases, they achieve registration.(4) To improve the speed of registration, linear difference and center differentialwere combined in this paper. At the same time multiresolution strategy wasused to achieve the algorithm implementation which improved the algorithm execution speed effectively, and avoid the generation of local extreme values,guarantee the reliability and accuracy of the registration.The experiments show that the method has some superiority and it has the strongrobustness.