Demons algorithm is an intensity-based fully automatic deformable image registration method.Its result is liable to be affected by the inconsistency of the intensity between the images，and its iterative convergence process is relatively slow.In this paper，a novel pretreatment approach of intensity normalization based on polynomial least squares fitting has been implemented to match the intensity between the images，and an accelerating algorithm for adaptively adjusting the deformation force of the floating image during the iterative process has been designed to speed up the process of convergence.This improved algorithm，the original algorithm and other improved algorithms have been applied to mammogram registration，and the results of the comparison show that the new method can obtain more accurate results with faster speed.The registered image which enhances genuine alterations in breast tissue would benefit the early detection of breast cancer.