Overview

In this project, we take images from the Prokudin-Gorskii photo collection and process them. Each pre-processed image consists of three color channel images (Blue, Green, and Red) which we align and combine to create a single color image.

Methodologies

Each image is split into RGB color channels and G and R channels are aligned to the B channel. For .jpg files, the offsets are calculated using an exhaustive grid search over a 15-pixel-by-15-pixel window, and the horizontal and vertical offsets are found with the least sum of the squared difference in relation to B. About 10% of pixels are cropped from all sides of each color channel so that the intensity-based metric will only account for information in the image. The results from this change (see below) significantly increase the accuracy of the offsets.

For .tif files, I used an image pyramid of 4 to 5 layers (determined by input size) so that the coarsest layer would have comparable input size as .jpg files. Each layer sizing down from the previous layer by a factor a two. The offsets are calculated iteratively from the coarsest layer to the original layer, and at each step, total offsets are updated and accounted for before finding the best offset in the next layer. This method produces very nice results on all images including the additional ones downloaded from the Prokudin-Gorskii collection.

Effects of Cropping Edges

Colorize without Cropping	Colorize with Cropping

Example Pictures

Name and Offsets [x, y]	Results
Cathedral G:[2, 5] R: [3, 12]
Tobolsk G:[3, 3] R: [3, 6]
Monastery G:[2, -3] R: [2, 3]
Harvesters G:[17, 60] R: [14, 124]
Emir G:[24, 49] R: [56, 104]
Icon G:[17, 41] R: [23, 90]
Lady G:[9, 52] R: [12, 112]
Melons G:[10, 82] R: [13, 178]
Onion Church G:[26, 52] R: [36, 108]
Three Generations G:[14, 53] R: [11, 112]
Self Portrait G:[29, 79] R: [37, 176]
Train G:[6, 42] R: [32, 87]
Village G:[12, 64] R: [22, 138]
Workshop G:[0, 52] R: [-12, 104]

Additional Pictures

This set of pictures are downloaded from the Prokudin-Gorskii collection.

Name and Offsets [x, y]	Results
House G:[20, 9] R: [40, 36]
Deviatiny G:[13, 61] R: [13, 156]
Slavianka River G:[-6, -31] R: [-14, -43]
Milan Cathedral G:[-9, 48] R: [-15, 111]
Lugano G:[-16, 41] R: [-29, 92]
Museum G:[20, 78] R: [25, 152]
toolkit.jpg G:[2, 66] R: [-7, 133]