Research On Image Registration Algorithm Based On B-splines Level Set

Image registration is a technique for processing two or several images obtained under different time periods,different viewing angles and different sensors and looking for an optimal spatial transformation map.It is a basic and important research topic in the field of image processing and computer vision.It has a wide range of applications in remote sensing image,image segmentation,target location,medical image analysis and other industrial applications.The traditional B-spline based Free-Form Deformation(FFD)registration method embeds the image into the mesh defined by the B-spline and evolves the image by deforming the B-spline object.Although a good registration effect is obtained in an image with small deformation or small local distortion,when there is irregular large deformation or partial severe distortion in the image,the method cannot simulate the deformation field and obtain an ideal registration result..The current B-spline algorithm is less robust to large deformation and topologically altered images,and the control point mesh cannot accurately capture the topological variation of the deformation field.Based on this,based on the B-spline registration algorithm based on linear elastic model,combined with the deformation field similar to the differential homeomorphic Demons algorithm,the robustness of the irregular large deformation image is enhanced while maintaining the topology.The gradient calculation of the B-spline registration method is derived from the integer-order differential and is partially ineffective for the gray-scale uniform and weak texture problems existing in the image.The main work of this paper is as follows:(1)The B-spline based free deformation registration algorithm has proven to be an effective method for solving non-rigid registration,but it is not ideal for large deformation registration.In response to this problem,when performing optimization at the B-spline control point,the displacement field is explicitly calculated for each registration step in a manner similar to the optical flow method or the differential homeomorphic Demons.The grayscale gradient is calculated directly from the derivative of the basis function.The explicit displacement field is projected onto the space of the B-spline transform using a composite update method to provide a smoothing effect for the deformation map,and image filtering techniques are used to efficiently calculate the gradient and B-spline coefficients,while helping to minimize computational complexity.The use of B-splines to represent image data and deformation is,in a sense,equally variable,with the potential to be more tightly integrated with software.(2)The existing gradient based on B-spline registration algorithm is derived from integer-order differential,which is easy to cause local invalidation for gray-scale uniform and weak texture image optimization process.B-spline registration algorithm and RL(Riemann-Liouville)Based on the combination of fractional differentials,a non-rigid registration algorithm combining RL fractional differential and B-spline level sets is proposed.Replacing the original B-spline level set gradient by the R-L fractional step can enhance the detail information of the image while enhancing the gradient information of the gray uniform and weak texture regions,thereby improving the image registration accuracy and speed.