New stopping criteria for iterative blind image deblurring based on residual whiteness measures

Blind image deblurring (BID) is a challenging ill-posed inverse problem. Most BID methods work by imposing some regularization on the unknown blur, and use iterative optimization schemes. Recently, a method was proposed that, although not requiring prior knowledge on the blurring filter, achieves state-of-the-art performance on a wide class of real-world BID problems. The drawback of that method is that the iterations have to be manually stopped. In this paper, we propose new stopping criteria for iterative BID algorithms. The rationale behind the proposed criteria is that if the blur filter is well estimated, the residual will be spectrally white, whereas with a wrong filter, the image estimate will exhibit structured artifacts, which are not white. Comprehensive experiments using the proposed criteria to stop the method mentioned in the previous paragraph show that it yields, on average, an ISNR only 0.16dB (~ 3%) below what is obtained by stopping the algorithm at the best ISNR (which, of course, can´t be done in practice).