Restoration of motion-blurred image by generalizedinverse method based on SVD

When a method based on linear model is applied for motion-blurred image restoration, its least square solution is the best linear unbiased estimator for the restoration. Because of the ill-conditioned degeneration of the image, this solution always diverges far from the original value in the case of noise jamming. In order to overcome the shortage, some subspaces not susceptible to noise were extracted by singular value decomposition (SVD) of degenerate matrix. A more robust inverse matrix was reconstructed on these spaces and it ensured the restored image had less distortion in a longer blurred length.