We propose a blind method to correct the optical aberrations caused by the point-spread function of the lens, without any prior on the lens or the camera to restore the image. Photograph taken with a Sony FE 35mm f/1.8 lens at maximal aperture mounted on a Sonyα6000 camera.

Abstract

The optics of any camera degrades the sharpness of photographs, which is a key visual quality criterion. This degradation is characterized by the point-spread function (PSF), which depends on the wavelengths of light and is variable across the imaging field. In this paper, we propose a two-step scheme to correct optical aberrations in a single raw or JPEG image, i.e., without any prior information on the camera or lens. First, we estimate local Gaussian blur kernels for overlapping patches and sharpen them with a non-blind deblurring technique. Based on the measurements of the PSFs of dozens of lenses, these blur kernels are modeled as RGB Gaussians defined by seven parameters. Second, we remove the remaining lateral chromatic aberrations (not contemplated in the first step) with a convolutional neural network, trained to minimize the red/green and blue/green residual images. Experiments on both synthetic and real images show that the combination of these two stages yields a fast state-of-the-art blind optical aberration compensation technique that competes with commercial non-blind algorithms.

Paper, code and online demo

Correspondence with Thomas Eboli.

Citation

            @inproceedings{eboli22fast,
              author       = {Thomas Eboli and
                              Jean{-}Michel Morel and
                              Gabriele Facciolo},
              title        = {Fast Two-Step Blind Optical Aberration Correction},
              booktitle    = {European Conference on Computer Vision},
              pages        = {693--708},
              publisher    = {Springer},
              year         = {2022}
            }
        

Method

The method removes the optical aberrations on the linear RGB color space (typically after demosaicking and before color and tone manipulations). It decomposes the problem into two simpler steps motivated by the structure of aberrations as showned in [Chang et al., TIP 2013]: (i) we deblur the edges with a Gaussian-like blind deblurring technique to get chromatic sharper edges, and (ii) we align with a new CNN the sharpened chromatic edges by operating on the pairs of green/red and green blue channels to render a realistic achromatic one. The full method is ligthweight, generalizes to new images and totally differentiable to be included in new ISP pipelines.

Results

Comparison on real JPEG images with two state-of-the-art techniques. The images are "linearized" by inversed gamma curve.

Original

Yue et al.

Li et al.

Ours

Comparisons on real raw images with the commercial solution DxO PhotoLab. The denoising and demosaicking are those of PhotoLab. Only the aberration processing module changes.

Original

DxO

Ours

Original

DxO

Ours

References

• Yue et al. Blind optical aberration correction by exploring geometric and visual priors. CVPR 2015
• Li et al. Universal and flexible optical aber- ration correction using deep-prior based deconvolution. ICCV 2021
• DxO PhotoLab 5