Computational Photography Project 1: Images of the Russian Empire

Image Alignment

In order to align Prokudin-Gorskii's glass plate images, I first used normalized cross-correlation to find the offset of one image to another. The idea was to take a central subregion of each of the images, do normalized cross-correlation and find coordinates of the peak, which is where the subregions are best correlated. Because using this offset some images were misaligned, I decided to use SSD which did measure correctly how well images match against each other. For the multi-scale implementation, I have used a gaussian pyramid, which recursively aligned the large scale images.