Project 1

Sergei Mikhailovich Prokudin-Gorskii had a simple idea in 1907 to take color photographs. He would take 3 different images, each with different color filters for red, green, and blue, that would produce a black and white images representing those colors. These images would later then be combined to create a full color image. There was no way to print these at the time, but he hoped that one day someone would create a projector to show these images off to Russian kids. In this project, we reconstruct the images taken by Prokudin-Gorskii using image processing techniques.

The images come on a plate with the 3 exposures, blue green and red from top to bottom. These are cut into thirds to separate each channel. The blue and red channel are aligned to the green channel using a technique called Sum of Squared Differences (SSD). Green was chosen as the base channel empirically - all channels were used as a base and green was chosen because it produced the best results across all images.

Alignments are searched using a pyramid technique. First, images are cut in half sequentially until they are below a certain size. All of these are saved. One by one, we align successively larger images in the sequence, starting from the offset calculated in the previous pyramid level. This allows for quick searching because a 1 pixel disparity in a lower pyramid level accounts for a 2 pixel disparity in the next highest level. Each level is searching in a range of [-15, 15]. When calculating difference using SSD, only the interior 50% of the image is considered to account for the fairly gnarly borders.

Once all the alignments are calculated, the final image is constructed by placing each channel in its respected channel, offset by the final calculated value.