This paper proposes a method for motion magnification that accelerates phase extraction using a Complex Steerable Pyramid (CSP). Motion magnification is a task to visualize minute fluctuations in a video that are difficult for humans to perceive. The conventional method amplifies the phase change of the CSP between frames in a video to emphasize the small fluctuations. These responses are calculated by the product of the transfer function of the complex steerable filters and the frequency components of each frame. However, the fast Fourier transform and its inverse transform required for this computation occupies about 30 % of the computation time, and is a computational bottleneck. We have accelerated the conventional method by generating the CSP by convolution in the spatial domain focusing on the hierarchical structure of image pyramids.