Blind image deblurring with noise-robust kernel estimation

Chanseok Lee*, Jeongsol Kim, Seungmin Lee, Jaehwang Jung, Yunje Cho, Taejoong Kim, Taeyong Jo, Myungjun Lee, Mooseok Jang* ;

"Blind deblurring is an ill-posed inverse problem involving the retrieval of a clear image and blur kernel from a single blurry image. The challenge arises considerably when strong noise, where its level remains unknown, is introduced. Existing blind deblurring methods are highly susceptible to noise due to overfitting and disturbances in the solution space. Here, we propose a blind deblurring method based on a noise-robust kernel estimation function and deep image prior (DIP). Specifically, the proposed kernel estimation function effectively estimates the blur kernel even for strongly noisy blurry images given a clear image and optimal condition. Therefore, DIP is adopted for the generation of a clear image to leverage its natural image prior. Additionally, the multiple kernel estimation scheme is designed to address a wide range of unknown noise levels. Extensive experimental studies, including simulated images and real-world examples, demonstrate the superior deblurring performance of the proposed method. The official code is uploaded in https://github.com/csleemooo/BD_noise_robust_kernel_estimation."