Robust image restoration for rotary motion degradations and the motion parameter identification

This paper constructs a frequency domain framework to deblur rotary degradation. The proposed method quantifies the rotary blur mechanism as frequency response and yields mean-square optimal estimation for original pixels under different noise levels. Then we present two methods to identify motion parameters. To detect correct rotary center, we propose a novel interval-searching method that works on the degraded pixel spectrum. The method is robust to noise. And a divide-conquer method is presented to iteratively calibrate the blur angle. Experimental results illustrate the frequency method outperforms spatial algorithms by up to 0.5 _~ 4 dB in Peak-Signal-Noise-Ratio (PSNR) and prove the methods for parameter identifications effective.