Abstract :
In the classical image restoration problem, the point spread function (PSF) is known. However, in many cases the PSF is unknown. In this case, image restoration is called blind image restoration. Blind image restoration algorithms can be divided into two categories: First, we estimate PSF through the image feature, and then the image is restored with the estimated PSF; second, the identification of PSF and image restoration are combined. The EM algorithm belongs to the latter. But when there is large noise, the result of the EM algorithm is poor. The nonlocal means (NL-means) algorithm has excellent denoising ability. In this article, first of all we use the NL-means algorithm to filter the noisy image. Then we use the result as input to the EM algorithm, which can solve the instability problem of EM algorithm caused by large noise. Experimental results show that this method can recover the original image and the PSF accurately.
Keywords :
expectation-maximisation algorithm; feature extraction; image denoising; image restoration; optical transfer function; EM algorithm; NL-means algorithm; PSF estimation; blind image restoration algorithm; classical image restoration problem; denoising ability; image features; nonlocal means algorithm; point spread function; Deconvolution; Degradation; Filtering algorithms; Image restoration; Noise measurement; PSNR;
