An Efficient Motion Blurred Image Restoration Scheme based on Frequency Domain Estimation

An image restoration scheme using frequency domain estimation to counter the motion blur effect is presented. The blurring process is characterized as a point spread function (PSF), which is further decomposed to a defocusing kernel and a motion kernel. The defocusing kernel, modeled as a Gaussian function, is estimated in the spatial domain while the motion kernel estimation is performed in the frequency domain. A motion-blurred image is observed to feature an ellipse-shaped spectral pattern and its orientation and the two axis lengths are closely matched to the parameters of motion kernel. After PSF estimation, a Wiener filter is employed for image restoration. Experimental results show that the proposed scheme can tackle the motion blur and as well as the defocusing blur effectively. The PSNR enhancements for the test bench images range from 2 to 4dB, which are better than the improvements achieved by well recognized tools such as Photoshop and SmartDeblur.