Multiscale guided deblurring: Chromatic aberration correction in color and near-infrared imaging

Chromatic aberration, caused by photographic lens imperfections, results in the image of only one spectral channel being sharp, while the other channels are blurred depending on their wavelengths difference with the sharp channel. We study chromatic aberration for a system that jointly records color and near-infrared (NIR) images on a single sensor. Chromatic aberration in such a system leads to a blurred NIR image when the color image is in-focus and sharp. We propose an algorithm that deblurs the NIR image using the gradients of the sharp color image, as both scene representations are generally similar. However, the details of these images often exhibit significant differences due to varying scene reflection and absorption in the corresponding bands. To account for this, we compute the correlation between color and NIR gradients, and use the gradients of the color image in reconstructing NIR only where the gradients are highly correlated. We propose a multiscale scheme that gradually deblurs NIR and accurately computes similarities between color and NIR gradients. Experimental results show that our algorithm recovers details of NIR without producing visible artifacts.