High quality single picture motion to blur __ image processing

Absrtact: We propose an algorithm to remove motion blur from a single image. Our method uses a uniform probabilistic model for estimating the convolution kernel and for the clear image in the process of fuzzy image calculation. We analyze the causes of the artificial traces that usually exist in the current blur method, and then introduce some new terms in our probabilistic model. These terms include the spatial stochastic model of fuzzy image noise, and the new local smoothing priori knowledge. By contrast constraint, even low contrast fuzzy images can reduce the artificial ringing effect. Finally, we describe an effective optimization scheme, through alternating estimation of fuzzy kernel and clear image restoration process until convergence. After these steps, we can get a high quality and clear image in a low computational complexity time. Our method produces image quality that is equivalent to the effect of a clear picture generated by multiple blurred pictures, and the latter method requires additional hardware resources.

Note: This article is my 10 years of translation of the Chinese University of Hong Kong Jayaya published in SIGGRAPH ASIA 2008 article, a lot of local translation is not good, please forgive me.

Please download the translated version from here.

Click to open the link

Original Download Address: http://appsrv.cse.cuhk.edu.hk/~leojia/projects/motion_deblurring/

Jayaya's paper is the classical literature of blind deconvolution algorithm, which involves many methods of modern image processing. A lot of the fast algorithms that came up later were improvements to him. The disadvantage of this algorithm is too slow, its processing effect is very good.