In deep space exploration, the star images obtained with the star sensor will suffer from severe motion blur due to multiple compound motions, and the signal-to-noise ratio is low. The Sureshrink wavelet method is proposed to denoise and it can effectively preserve the star image edge information and gray information. Meanwhile, a deconvolution model of star image based on the maximum posteriori estimation was proposed according to the gradient of star image obeying Laplace distribution, and then it was utilized to recover the blurred star images. Simulation results show that the extraction accuracy is dramatically improved after deblurring, the proposed method generates less ringing effect compared with Wiener filter and presents better performance in terms of extracting speed and accuracy compared with Richardson-Lucy filter, and the centroid accuracy is better than 0.1 pixels, which satisfies the accuracy and speed requirements of the dynamic star sensor.