Iterative blind image restoration with ringing reduction

Abstract

Image restoration is one of the most important problems in image processing. It is extremely difficult to obtain the accurate PSF (Point Spread Function) of the degraded image captured from imaging system, so the method of iterative blind deconvolution which is superior to the other ones is adopted. In blind image restoration, the estimation of initial PSF plays a significant role. Meanwhile, with the increase of the number of iteration, ringing ripples may appear at boundary and the regions where gray values vary severely, which affects the quality of the restored image. Therefore, estimation of initial PSF and ringing reduction are two primary problems in blind iterative image deconvolution. A novel approach combined the estimation of initial PSF with ringing suppression in blind iterative restoration is proposed. First, a sophisticated variational Bayesian inference algorithm with natural image statistics is used to estimate initial PSF. Cyclic boundary method is applied to suppress the ringing artifact at the boundary. As to the ringing in areas near edges, spatial weighted matrix is adopted, because spatial weighted matrix can impose an adaptive local constraint on restoration and smoothing to achieve better results. Finally, image quality assessment is used to evaluate the restored image, which also could agree with human visual system. Experiments of simulation and real images show that this approach can restore boundary and edge areas well and reduce the noise in smooth regions, which means that this method can suppress ringing ripples and preserve more details validly.