The goal of this project was to produce colored images from the digitized Prokudin-Gorskii glass plate images using image processing techniques. The first step was to extract the three color channel images and then align them by one of two processes:
For smaller photos (.jpg), I searched over a [15, 15] window of displacement between two photos and took the one with the least sum squared differences. Then the 3 color channel images would be aligned by the produced offsets. In addition, I had to crop out the borders to improve the SSD so I cropped out .15 of the photo from each side.
|
|
|
|
For bigger photos (.tif), the higher resolution made the previous approach too inefficient, so I used the pyramid method instead. I recursively called the previous align method on a scaled version of the photo with a [-15, 15] window, and then began scaling the image back and calling on the align method on it with a smaller window. At first I was encountered a bug that I was stuck on for a while before realizing that I needed to multiply the offset by the factor I was scaling it by (in this case 2) to keep my results accurate.
|
|
|
|
|
|
|
|
I also encountered a problem with emir.tif (from piazza I could see that some other people did as well) because it had a dominant presence of blue. To fix this, I aligned it images against the green channel instead of the blue channel.
|
|
|