Objective:To evaluate the value of 2D-3D registration method which combines improved mutual information and image pyramid.Methods:The continuous image representation of the cubic B-spline curve and Parzen histogram estimation were fused into the algorithm. The chest was used as the research object. The reconstruction of the orthogonal X-ray image generated by the radiographic image and the image after a certain transformation with itself were used for registration experiments to study the registration accuracy and time.Results:After 50 sets of controlled experiments, compared with the traditional registration method, the displacement accuracy of this method in the X and Y directions was improved by 53.39% and 21.33%, and the registration time was shortened by 91.93%. Compared with the modified algorithms in recent years, the displacement accuracy of the improved algorithm in the X and Y directions was increased by 17.65% and 13.79%. And the registration time was further increased by 19.64%.Conclusions:This method can effectively improve the registration accuracy and efficiency of 2D and 3D images, and both meet the requirements of image registration within 2 mm during surgery. The high efficiency and accuracy of this method provide beneficial information for clinical diagnosis and radiotherapy automation, which also lays the foundation for tumor position error correction and automatic positioning of medical robotic arms.