Kernel estimate for image restoration using blind deconvolution
Alexander Tselousov, Sergei Umnyashkin · 2017
In most cases image distortions modelled by convolution and additive white noise have unknown model parameters, such as convolution kernel (point spread function - PSF) and noise power. Different methods of blind deconvolution which iteratively approximate PSF use some initial kernel estimation; their performance is sufficiently dependent on the precision of that estimate. Modelling initial PSF as a line segment we propose a hybrid method of its estimate which is based both on cepstral and gradient field analyses of distorted image. The proposed method, combined with spectral estimation of noise power, results in better performance of blind deconvolution image restoration which in our experiments was based on Bayes approach and Lucy-Richardson engine. Both visual and measured quality of restored images becomes better, computational load gets lower.