Automatic Digicromatography: Colorizing the Images of the Russian Empire

A hundred years ago, before color photography existed, a Russian scholar ProkudinGorskii attempted to create color photographs by taking three black and white exposures of the same scene through three filters: blue, green and red, to be displayed using a special triple lens projector. Today, 1,902 three-channel negatives of these photographs survive and are available in digital format on the Library of Congress website. A color image can be reconstructed from a three-channel negative by manual alignment and careful color adjustment, termed digicromatography. However, the manual process is too time-consuming to be used on the entire collection, so an automated approach is needed. In this work we have investigated several ways of automating digichromatography. After the initial automatic color alignment using the Gaussian pyramid and affine warping, there are three types of image artifacts that must be fixed: 1) local level artifacts such as extreme red/green/blue spots from dirt particles on the negatives; 2) mid-level artifacts such as gradual region color change; and 3) global artifacts such as unnatural or bleak color in the whole picture. We have applied different image processing techniques for each of these types of artifacts. Our fully automatic approach demonstrates promising results on many of the images. (a) (b) Figure 1: shows a negative (a) from Prokudin Gorskii’s collection, and a manually composite image (b) of the negative by Alex Gridenko This demonstrates the ideal goal of our work, which is to create a program that takes a negative and produce a beautiful image like in (b)