Automatic Tuning of Gaussian Filter for Image Vignetting Correction

Image vignetting is a common optical artefact characterised by a gradual reduction in brightness towards the edges of an image. It degrades image quality and compromises radiometric accuracy, affecting a wide range of imaging applications. When vignetting cannot be avoided during acquisition, a computational correction is required. Existing methods often rely on Gaussian filtering of a flat-field image; however, selecting the appropriate filter parameters—particularly the standard deviation—remains challenging and is largely subjective. To address this issue, this paper presents the Gaussian filter with auto-tuned sigma (GFATS) method. This method leverages an optimisation framework to automatically tune filter parameters. This is achieved by aligning the filtered output with a specified polynomial model of the vignetting profile near the optical centre of the captured flat-field image. The tuned filter is then applied to the same image to derive the vignetting correction matrix. The proposed method was evaluated against established model-based correction methods across different lens–camera systems using objective quantitative measures. The results demonstrate that GFATS provides a more accurate vignetting estimation and improves the brightness uniformity of corrected images compared to existing approaches. It also effectively mitigates the overfitting issue inherent to standard low-pass filtering or smoothing methods. Notably, a single consistent parameter set was found to provide a reliable and stable performance across all tested lens–camera systems, demonstrating the method’s versatility and strong practical applicability.