This project was about recovering the color photographs of Russia made by Prokudin-Gorskii. This was especially interesting for me, because Russia is my home country. Also I love taking photographs, so it's fascinating for me to know that someone was already taking color photographs over a century ago.
Approach:
My approach was pretty much what was in the instruction: I was shifting the Green and Red color channels over the Blue one using a for loop, after each step I calculated the Sum of square differences and put it in the matrix. Then I was searching for the minimum element in the matrix, and shifted the channel image to the coodinates of that element. This usually shows OK results, however on some images there are artifacts. I usually used 15 or 30 pixels shifting. Probably easier solution was to just find the minimum Sum and not to fill the matrix, but I like to see the data. To add some of the bells and whistles, to remove the artifacts, I cropped the image matrix by 30 pixels, this is the area that usually contains artifacts and bad data. The recursive pyramid algorithm I usually use on 3 or 4 levels.
Problems:
The hardest part was to implement the recursive algorithm. Initially, it took like a minute to produce a bad image. But eventually I improved the quality and the time to around 11-17 seconds for a large tiff.
Assignment Pictures:
|
|
|
|
|
|
|
|
|
The boat picture does not look good, probably because it required a bigger window of displacements. Images from the collection:
Some examples of TIFF Pictures:
Displacements:
00106v.jpg - 9,1
00757v.jpg - 5,1
00888v.jpg - 10,1
00889v.jpg - 5,1
00907v.jpg - 5,1
00911v.jpg - 7,1
01031v.jpg - 4,1
01657v.jpg - 12,1
01880v.jpg - 13,3