Truncated Cauchy Loss for Outlier-Robust Blind Image Deblurring
Xiaopan Li, Shiqian Wu, Qi Wu · 2024
In complex imaging environments, captured images often contain outliers like impulse noise, Cauchy noise, and saturated pixels, posing significant challenges for image deblurring. While existing blind deblurring methods can address outliers to a certain extent, the incorporation of intricate operations tends to amplify the complexity of the deblurring process. In light of these challenges, we present a simple yet robust blind deblurring algorithm for handling images with outliers. To alleviate the effect of outliers on kernel estimation, we introduce a truncated Cauchy loss function, which effectively suppresses outliers by truncating significant errors. Furthermore, we employ a statistics-based approach for outlier detection, eliminating the need for heuristic methods. The proposed approach is shown to be easily implementable and extendable for non-blind deblurring. Experimental results demonstrate that the proposed algorithm outperforms state-of-the-art methods for both blurred images with outliers and those without.