Automatic photometric restoration of historical photographic negatives

Abstract

The majority of early photographs were captured on acetate-based film. However, it has been determined that these negatives will deteriorate beyond repair even with proper conservation and no suitable restoration method is available without physically altering each negative. In this paper, we present an automatic method to remove various nonlinear illumination distortions caused by deteriorating photographic support material. First, using a High-Dynamic Range structured-light scanning method, a 2D Gaussian model for light transmission is estimated for each pixel of the negative image. Estimated amplitude at each pixel provides an accurate model of light transmission, but also includes regions of lower transmission caused by damaged areas. Principal Component Analysis is then used to estimate the photometric error and effectively restore the original illumination information of the negative. Using both the shift in the Gaussian light stripes between pixels and their variations in standard deviation, a 3D surface estimate is calculated. Experiments of real historical negatives show promising results for widespread implementation in memory institutions.