Blind aberration correction for light field photography

Aberration correction is critical for obtaining sharp images but remains a challenging task. Owing to its ability to record both spatial and angular information of light rays, light field imaging is a powerful method to measure and correct optical aberrations. However, current methods need extensive calibrations to obtain prior information about the camera, which is restrictive in real-world applications. In this work, we propose a two-stage blind aberration correction method for light field imaging, which leverages self-supervised learning for general blind aberration correction and low-rank approximation to exploit the specific correlations of light fields to further abate aberrations. We demonstrated experimentally the superiority of our method over current state-of-the-art.