Abstract:  Image registration is one of the basic tasks of image processing, which is widely used in military, remote sensing, medicine, computer vision, pattern recognition and other fields. image registration is to make an image
consistent with another image on the corresponding point, surface, or pixel values to using geometric transform in the same space，which will be a variety of modes of image information fusion into a new image. Multiple images
from the same object arc taken by the same or different sensors, at different times or different points of view, so that in these images there arc some differences，we must take image registration before using these images to do
research. According to the deformation characteristics of the image classification, image registration can be divided into rigid registration and norrrigid registration categories. For the thre-dimensional image，rigid registration is to find a
six-degree of freedom (three rotational，three translational) of the transformation, making floating point image is mapped to the corresponding point of the reference images. Norrrigid registration is to regulate the
rigid deformation between images，such as image stretching deformation, bending deformation complex noe tensional anc1 deformation and so on, In many cases rigid registration can not meet the needs of the actual image, because the
nature of a lot of images deformations arc norrrigid，which need use norrrigid registration method to registration. Noe methods. rigid registration method is calculating the deformation field offset through parametric or norr
Parametric method is using basis functions to express the main deformation field，which contain Iaramctnc thin plat spline function, Basirspline function. Norrparametric method is to directly calculate the offset of each pixel，such
as diffusion model，viscous fluid model，DEMON methods. Based on Yrsplinc registration algorithm can control the local deformation, which affects only the local deformation of the field by changing the control points. Free
deformation model which has a good ability to simulate the local deformation, also has the unrestricted freedom and the advantage of efficient computation, is the best option of simulations of the local norrrigid deformation for image
registration. Image Norrrigid Registration is the hotspot and difficulty of image processing research, its parametric models arc often transformed into unconstrained optimization problem which is always the research emphasis of the
theoretical research. its time-consuming for matrix’s calculating and storing in per iterative when solving the problem. In this paper we provide a norrrigid registration method，which is based on Basirspline free form deform
model.This model can make only partial deform, and has implicdly added the smoothness constraint.This method utilizes limited memory Broyden-Fletcher}oldfarb-Shanno optimal method for solving the cost function, which avoid
calculating the Hessian matrix and the its mvcrsc matnx.This optimal method reduces computational complexity and memory method capacitance, and make cost time shorten much..The experiments prove this norrrigid registration
is not only of high efficicncy,but also of good effect in registration.