Three real instances of blur: camera shake from the RealBlur dataset (scene 118, image 2), a subset of the local calibrated optical aberrations from the Canon EF24mm f/1.4L USM opened at f/2.8, and the result of the ×2 multi-frame super-resolution. For shake the light streaks suggest the blur is roughly the same everywhere, thus patches can be sampled uniformly. The same holds for fusion blur where the fused image globally lacks of sharpness. The aberrations are all unique but roughly follow a central symmetry, thus easy to sample.

Abstract

Blurry images usually exhibit similar blur at various locations across the image domain, a property barely captured in nowadays blind deblurring neural networks. We show that when extracting patches of similar underlying blur is possible, jointly processing the stack of patches yields superior accuracy than handling them separately. Our collaborative scheme is implemented in a neural architecture with a pooling layer on the stack dimension. We present three practical patch extraction strategies for image sharpening, camera shake removal and optical aberration correction, and validate the proposed approach on both synthetic and real-world benchmarks. For each blur instance, the proposed collaborative strategy yields significant quantitative and qualitative improvements.

Citation

            @article{eboli23collaborative,
              author       = {Thomas Eboli and Jean{-}Michel Morel and Gabriele Facciolo},
              title        = {Collaborative Blind Image Deblurring},
              eprint       = {2305.16034},
              archivePrefix= {arXiv},
              year         = {2023}
            }