Colorizing the Prokudin-Gorskii photo collection

Introduction

The goal of this assignment is to take the digitized Prokudin-Gorskii glass plate images, and using image processing techniques, automatically produce a color image with as few visual artifacts as possible. In order to do this, I extract the three color channel images, place them on top of each other, and align them so that they form a single RGB color image.

Methodology

Input: Each processed input consists of 3 glass plate images, corresponding to read, green and blue channels, as follows:
Align: The most naive approach is to simply stack the channels. The result would be blurry. A better approach is to anchor a channel and try to align the other two channels to this "anchor" channel. To align, we search over a window of displacements, and evaluate some "distance function" for the aligned image and the anchor image. Two popular distance functions are: Crop: We then crop 15% from each side to remove the black edges. More advanced cropping techniques including running edge detector then cropping areas with continuous large area of edges.

Large Images

For large images, it is extremely slow and inefficient to perform the window method over a large set of displacements. Therefore we apply the pyramid method. We downscale the image, and use a small window size to align the coarse image; with that alignment, we double the scale of the image, so on and so forth. For each scale, we only use a small window; however we are moving to a more accurate alignment gradually over the scales. An image illustration is below:

Advanced Features

In the case of Emir of Bukhara, the channels do not actually have the same brightness values. This posts a significant challenge, since RGB intensity does not provide a good feature. Running our naive pyramid implement would yield bad result

Instead of looking at RGB feature, I used canny edge detector to get edge matrix for each of the channel glass. Using edge information as feature, I performed the same pyramid matching method on the edge matrices, and obtained the best displacements, and using them to align. As can be observed, now the alignment is much better.

Auto Constrast

To make the color more contrasting, we look at the exposure histogram of the image, and try to make the distribution as uniform as possible.