15-463: Computational Photography

Project 1: Images of the Russian Empire

Eugenio Dominguez

For this project the task was to find and algorithm that automatically aligns the three black and white RGB channels, taken by Sergei Mikhailovich Prokudin-Gorskii, to produce color photographs. My algorithm works in the folowing way:

- Divide the original three channel photograph in three equal parts and use each part as one channel of the color image.
- Move the red channel over the blue channel one pixel in one direction
- Crop the borders of the picture (5% of the picture width)
- Calculate the Sum of Squared Differences between the red and blue channels
- Repeat the last 3 steps for all directions in a window of displacements provided by the user (usually 15 pixels for the small images and 120 for the large ones)
- Choose the displacement with the minimum SSD
- Do the same for the green channel

For the large images, I implemented an image pyramide for faster processing.

Here the results in the sample images. On the left is the image without processing and on the right the image after the automatic displacement.














And my results for the larger images.





Here the algorithm failed. Most probably because the large blue area in the center of the image, that produces a big difference in brightness between the blue and the other channels, so the aligned image has a big SSD.









An two other images found in the Library of Congress