In this paper, a fast global motion estimation algorithm is proposed to speed up the processing of microscopic image sequence. A translation model based on non-linear density estimation is adopted by the proposed algorithm. To speed up the process, three-level pyramid is used. Noise detection and feature pixel extraction run simultaneously in each pyramid level. Then the motion parameters is estimated only with the feature pixels. The experimental results show that the computational complexity of the proposed algorithm is reduced much while estimation accuracy is maintained compared with the MPEG-4 algorithm.