Although affine model has more ability to describe the motion field

it's very difficult to be used for it's high computational complexity.In this paper

a novel fast algorithm is proposed which is derived from traditional phase-correlation motion estimation method.In FFT frequency domain

the phase shift indicates the translation components of affine motion and the spectrum magnitude indicates the linear components of affine motion.So we utilize this FFT spectrum characteristic to separate the affine parameters estimation to translation components and linear components.The computational complexity is decreased dramatically.For improving the robust of linear component estimation

we use an affine-invariant feature to calculate the correlative points between magnitude spectrums.Simulation indicates that our method can improve the estimation result greatly with sustainable computational complexity.