Efficient motion blur parameters estimation under noisy conditions

Image restoration of noisy motion blurred images is a challenging problem in image processing. The quality of restoration of such images depends on the perfect estimation of blur parameters. In this paper, an algorithm is proposed to estimate linear motion blur parameters such as blur length and motion direction under noisy conditions. In the proposed algorithm, pre-processing of the noisy image requires appropriate windowing and subsequently dual Fourier transform of the image. Blur angle is estimated accurately using Radon Transform on a specific bit plane image of the dual Fourier transformed image. For accurate estimation of blur length in presence of noise, the image is processed in bispectrum domain as the bispectrum suppresses additive Gaussian noise. A robust novel exponential model is proposed based on curve fitting that represents the behavior of blur length in bispectrum domain. The maximum error in the estimated blur angle and the blur length is less than 1.6° and 2 pixels respectively in presence of additive Gaussian noise of variance up to 150. Experimental results are also compared with the previous methods to demonstrate the superior performance of the proposed algorithm.