An end-to-end blurred image restoration algorithm based on CycleGAN is proposed to solve the complicated problems of obtaining real paired data sets of moving blurred images and simulating real blurred images with manually generated blurred images. In CycleGAN, only using the loss of average absolute error (L1) as the reconstruction loss wm lead to the problems of artifact, color distortion and unclear image in the restored image. The problem of image artifact and unclear image can be solved by introducing the loss of structural similarity measure (SSIM) and the color distortion can be solved by the loss of color saturation. The experimental results show that compared with several deblurring algorithms with better restoration results, the restoration results produced by the algorithm in this paper are significantly improved in qualitative and quantitative evaluation.