| Image registration is one of the basic techniques in image processing,and widely used in many fields like medical image analysis,computer vision and pattern recognition,etc.,and has been regarded as a hot research field.The cost function of the classic free-form deformation(FFD)based on multi-level B-spline consists of a similarity term and a smooth constraint term,which makes the FFD model perform well in image registration with little deformation or local distortion,but sometimes,can't better simulate transformation field of moving images with large elastic deformation or large local distortion.In this paper,in order to increase the registration accuracy and robustness,a new method based on the classic hierarchical FFD is proposed.Since the smooth constraint term has a significant influence on registration accuracy and its coefficient is hard to control in the cost function of the original hierarchical B-spline based FFD.Compared with the classic cost function,A L2-regularization term,namely the Frobenius norm of control points' matrix,is introduced in the new cost function of the proposed model.By coordinating the coefficients of this regularization term and the smooth constraint term,this novel L2-regularizd FFD model is able to solve the problem of low registration accuracy caused by strong smooth constraint while preserving the image's topological structure.The introduced Lz-regularization term is able to impose a spatial constraint on the control lattices'transformation field,and an over registration problem is avoided in a certain degree,so it can register the image with local larger distortion.In addition,L2 Norm has the good property of making optimization faster and more stable,which can improve the image registration robustness.By conducting a series of registration experiments of natural images and medical images,the new method has an obvious advantage over the classic hierarchical FFD model in registration accuracy measured by mean square error(MSE). |
PDF Full Text Request