Parameters identification of point spread function in noisy motion blurred image

The restoration of blurred image is highly dependent on accurate identification of blur parameters. For the unknown PSF parameters of uniform linear motion blur and the weakness of spectrum characteristics in noisy blurred image, a robust identification method is proposed based on frequency domain analysis with reference image. Specifically, a clear reference image chosen randomly is blurred with various blur length and directions of the same model, and then the spectrum correlation maximum between the blurred image and the image to be restored is analyzed to obtain the right model parameters. The experimental results show that there is no error estimation in the direction and the length when -90°≤α≤90° and 10≤L≤90 pixel for noise free blurred image, and for Gaussian noise blurred image with variance between 0.001 and 0.03, the worst case absolute estimation error is 2°in direction and 3 pixel in length. © 2013 by CESER Publications.