BBM 415 Assignment 1-Images of the Russian Empire Solution

Images of the Russian Empire: Colorizing the Prokudin-Gorskii Photo Collection

Figure 1: Blue, Green, Red negatives to RGB image.
Background
Sergei Mikhailovich Prokudin-Gorskii (1863-1944) [1], a Russian chemist and photographer, is considered as a pioneer in color photography. With Tsar’s special permission and funding, he traveled across the Russian Empire, and took color photographs of people, daily life, buildings, landscapes, etc., documenting the early 20th-century Russia. He produced these color photographs by using a simple but ingenious idea: recording three exposures of every scene onto a glass plate using a blue, a green, and a red filter. These negative plates is now in possession by the Library of Congress of U.S.A., and have been recently digitized. They are available on-line at [2].
Overview
The goal of this assignment is to automatically produce color images from the digitized Prokudin-Gorskii glass plate images by applying some simple image processing techniques. For that, you will need to extract the red/green/blue color channels from the digitized plate images, and find the best alignment so that they form an RGB color image with few visual artifacts. You can download a starter code and some sample plate images from the course webpage.
Details
Your program will take a glass plate image as input and produce a related color image as output. It should have the following structure:
1. Divide the input image into three equal parts corresponding to RGB channels.
2. Align the second and the third parts (G and R channels) to the first one (B channel).
The simplest strategy to align the parts is to perform an exhaustive search over a window of possible displacements (e.g. [-15,15] pixels), and take the displacement that gives the best matching score. Some possible matching metrics are as follows:
• Sum of squared differences (SSD)
• Normalized cross-correlation (NCC)
3. Improve the quality of the aligned image.
Some important remarks:
- Recall that the color filter order from top to bottom is BGR, not the typical RGB.
- Assume that the negatives are evenly divided into 3 plates (i.e., each plate is in exactly 1/3 of the negative).
- Assume that a simple x,y translation model is sufficient for proper alignment.
- In Python, you can easily shift a matrix by using roll function of numpy module.
- You can devise and utilize an algorithm to automatically crop the borders in order to eliminate the artifacts in the boundaries and obtain a better quality result.
What to Hand In
You are required to submit all your code along with a short report in HTML. For that purpose, prepare a folder containing
• README.txt (give a text file containing the details about your implementation, how to run your code, the organization of your code, functions etc.)
• code/ (directory containing all your code)
• html/ (directory containing all your html report, including your images)
• html/index.html (html report) Archieve this folder as b<studentNumber>.zip and submit to https://submit.cs.hacettepe.edu.tr. Your HMTL report should contain a brief overview of the problem, the details of your approach, and the results of your algorithm on at least 8 negative images with your comments. Show the results of all of the main steps (unaligned result, result after alignment, results of the enhancement). If your algorithm failed to give a satisfactory result on a particular image, provide a brief explanation of the reason(s).
Grading
The assignment will be graded out of 100:
• 0 (no implementation), 20 (an attempt at a solution), 40 (a partially correct solution), 60 (a correct solution).
• Good report preparation and your comments on problem and results you obtain will be graded over rest 40 points.
References
[1] http://en.wikipedia.org/wiki/Prokudin-Gorskii
[2] http://www.loc.gov/exhibits/empire/gorskii.html