Overview

In this project we will colorize images based on negative plates taken by Russian photographer Prokudin-Gorskii (1863-1944). Using monochrome images taken from 3 colored channels (RGB), we will use program to auto-align them and overlay them into a single colored image.

Part I: Exhaustive Search

This part of the project takes 3 colored negative plate images with a relatively small size (around 400 pixels per side) and calculates the best displacement to align the colors using Sum of Squared Difference (SSD) method over a range from -15 to 15 pixels. My program takes in as argument 2 color plate images: one to be shifted, the other as reference. The alignment function is ran twice to obtain red and blue displacements, both using green channel as reference image. These positional parameters are then used to produce a lined-up, colorized image.

Result

red (5, 2) blue (12, 3)

red (-3, 2) blue (3, 2)

red (3, 3) blue (7, 3)

Part II: Image Pyramid

For images with larger sizes and higher resolution, exhaustive search will become prohibitively expensive and slow. For this reason, an image pyramid is implemented to optimize the speed. My image pyramid function works by recursively resizing the image down to a smaller image, and then cropping it down to its central area for exhaustive searching for the best alignment for that "tier" of the pyramid. The cropping step is there to avoid taking into account noise at edges which may lead to less than optimal results. The program failed to align the emir image because the color intensity between the colored channels are so great that SSD cannot provide a satisfactory result.