Images of the Russian Empire: Colorizing the Prokudin Gorskii Collection

Including

A Scandalous Revelation of the Communist Nature Thereof

never before seen by anyone who lived to tell:

BEHOLD!

Ronit Slyper

Algorithm

In this assignment, slides from the Prokudin-Gorskii photo collection were automatically aligned and colorized. Matlab was used to read the photos, align them, and write them out again. GQView was used for scanning quickly through the results.

In the final code, I perform a crop on the entire image to get rid of the white boundary. The crop is automatic - it takes row and column averages of the entire image, and looks for the white border. For the large tif's, I crop another one to two hundred pixels for good measure. I tried cropping when I reached the point where my images were somewhat aligning, but the boundary was messing things up; cropping greatly improved the quality of the alignment.

Before beginning the alignment algorithm, I padded the image with a boundary of size user_bounds. This made subsequent code cleaner since when the green and red (unpadded (i.e. cropped)) layers were matched against the blue layer, they were guaranteed to lie entirely within the blue layer. The final metric I used for comparing layers ignored regions which were entirely zeros - something rare in the actual image data, so it was a reasonable hack.

Results with [x,y] translational alignment

JPG's

The example smaller jpg's, aligned, are shown here. The offsets are given in this matlab output. To compare the coefficients to other results, please note that the image is first cropped by the amounts listed in the file, then padded by user_bounds (inited at 15).

Pictures 00056v.jpg and 00498v.jpg were noticeably imperfect; the former because the slide is messy, and the latter because there is comparatively less structure in the image, so errors in border cropping swamp the alignment metric.

TIF's

matlab output
matlab output
matlab output.The camels aren't aligned right. I don't really like camels anyway. Re: whitespace + border effects

More TIF's from the website

matlab output

Bells & Whistles

Behold the leader of the Russian Soviet Socialist Republic, forever living and working among the workers and works of his great land. So pervasive is the might and breadth of the Glorious Red Channel, that Lenin is to be found among the people everywhere.

location	location	location	location A harsh critique of Communism it is, that Lenin aligns with a camel's ass.	location Most fun.
location	location	location Hardest.	Inspiration for the formation of the Secret Police.

Alignment was performed using normxcorr2 since the pictures were small enough. The alignment was performed on the image's red channel (for moral correctness; it sometimes gives odd results), then blended into the red channel. A solution to the puzzle of image blending satisfyingly was inspired by the paper "Feature Based Image Mosaicing". I used a blend mask with weights that increased towards the center of the mask.