Project 1

Summary

A sample source image

This project automatically reconstructs full color images from the Prokudin-Gorskii photo collection. The source images each have three separate images taken with different color filters: Blue, Green and Red. These images are given to a MATLAB program and are automatically split into three parts and aligned before being combined into a full color image.

Methods

Images were aligned in different ways depending on their size. Small images were aligned by performing a Normalized Cross-Correlation (NCC) operation using the Green channel as a base over a user-specified window. The largest value found in the window after the NCC operation is the offset for that channel. I used the window [-15, 15] for the small examples here. Template images were created by cropping off a 15 pixel border from the original channel image.

For larger images, an image pyramid speedup scheme was used. Each base image and template image were repeatedly downsampled to half-size until they were too small to perform any meaningful correlation. The NCC operation was then perfomed at this coarse level and the estimated offset was used in the next level down the pyramid as a starting point for a more limited search. I had the algorithm use a [-10, 10] window at each stage and created the template image by cropping off a 500 pixel border from the original channel image.

Results

All images used the Green channel as the base channel except where noted. Large images (the .tif files) will have an additional link to the full image in their descriptions.

File Name: 00029u.tif

Full Image

Red Offset: [53, 20]

Blue Offset: [-38, -15]

File Name: 00087u.tif

Full Image

Red Offset: [60, 16]

Blue Offset: [-47, -37]

File Name: 00106v.jpg

This image aligned better using the Red channel as the base. I suspect that the Green channel wasn't bright enough to get a good correlation with the other channels.

 

Green Offset: [10, 0]

Blue Offset: [5, 2]

File Name: 00128u.tif

Full Image

Red Offset: [18, 24]

Blue Offset: [-34, -24]

File Name: 00212u.tif

Full Image

Red Offset: [64, -5]

Blue Offset: [-47, 10]

File Name: 00458u.tif

Full Image

Red Offset: [44, 28]

Blue Offset: [-41, -5]

File Name: 00737u.tif

Full Image

Red Offset: [35, 9]

Blue Offset: [-14, -6]

File Name: 00757v.jpg

Red Offset: [4, 3]

Blue Offset: [-1, -2]

File Name: 00822u.tif

Full Image

Red Offset: [69, 9]

Blue Offset: [-56, -24]

File Name: 00888v.jpg

Red Offset: [7, 1]

Blue Offset: [-5, 1]

File Name: 00889v.jpg

Red Offset: [4, 2]

Blue Offset: [-1, 0]

File Name: 00892u.tif

Full Image

Red Offset: [27, 2]

Blue Offset: [-15, -1]

File Name: 00907v.jpg

Red Offset: [4, 0]

Blue Offset: [-1, 2]

File Name: 01031v.jpg

Red Offset: [4, 1]

Blue Offset: [0, 0]

File Name: 01043u.tif

Full Image

Red Offset: [27, 9]

Blue Offset: [17, -9]

File Name: 01047u.tif

Full Image

Red Offset: [48, 15]

Blue Offset: [-23, -19]

File Name: 01540u.tif

Full Image

Red Offset: [83, 9]

Blue Offset: [-81, -16]

File Name: 01657v.jpg

Red Offset: [7, 1]

Blue Offset: [-4, 0]

File Name: 01880v.jpg

Red Offset: [9, 3]

Blue Offset: [-5, -1]