Blind Image Restoration Based on Power Law
Honggang Sun, Fanhua Meng, Weizhe Gao, Shuang Yu, Zhe Yang, Yiduo Gao · 2023
The goal of blind image restoration is to recover the latent clean image from the observed blur version without complete prior information of the degraded factor. Commonly, blind image restoration employs the iterative algorithms, which are time-consuming, uncertain or divergent. This paper proposes a non-iterative blind image restoration algorithm based on power law (BRPL). BRPL focuses on a prevalent and significant class of point spread function (PSF) - Class G. The spectrum of most natural images follows the power law distribution. An isosceles model is designed to approximate the true image's spectrum. The PSF is estimated by comparing the spectrum of the blurry image with the reconstructed one. After that, the image is reconstructed by employing the obtained PSF and Wiener filter. The proposed algorithm is non-iterative, as a consequence it is ultrafast. The experiments show that BRPL estimates PSF more accurately and reduces ringing artifacts significantly compared with some existing algorithms. The quality of the restored images are improved remarkably. Moreover, BRPL works well in restored the color image.