Estimating vignetting function from a single image for image authentication

Vignetting is the phenomenon of reduced brightness in an image at the peripheral region compared to the central region. As patterns of vignetting are characteristics of lens models, they can be used to authenticate digital images for forensic analysis. In this paper, we describe a new method for model based single image vignetting estimation and correction. We use the statistical properties of natural images in the discrete derivative domains and formulate the vignetting estimation problem as a maximum likelihood estimation. We further provide a simple and efficient procedure for better initialization of the numerical optimization. Empirical evaluations of the proposed method using synthesized and real vignetted images show significant gain in both performance and running efficiency in correcting vignetting from digital images, and the estimated vignetting functions are shown to be effective in classifying different lens models.