15463 Computational Photography
15463 Computational Photography
Project #1 : Colorizing the Prokudin-Gorskii Collection
Varun Ramakrishna
Single-scale Alignment
The low-resolution pictures were aligned using the Normalized Cross-Correlation function. A search window of width 20 was used
for each image.
Image: '00163v.jpg' |
Image: '00270v.jpg' |
< |
|
Displacement: Red: x = 14 y = -1 Green: x = 7 y = 0 |
Displacement: Red: x = 6 y = -1 Green: x = 1 y = -1 |
Image: '00398v.jpg' |
Image: '00564v.jpg' |
|
|
Displacement: Red: x = 4 y = 3 Green: x = 2 y = 0 |
Displacement: Red: x = 5 y = 0 Green: x = 2 y = 0 |
Image: '01167v.jpg' |
Image: '31421v.jpg' |
|
|
Displacement: Red: x = 10 y = 3 Green: x = 5 y = 2 |
Displacement: Red: x = 7 y = 0 Green: x = 5 y = 0 |
Image: '00125v.jpg' |
Image: '00149v.jpg' |
|
|
Displacement: Red: x = -9 y = 1 Green: x = -4 y = 1 |
Displacement: Red: x = -11 y = 0 Green: x = -5 y = 0 |
Image: '00153v.jpg' |
Image: '00154v.jpg' |
|
|
Displacement: Red: x = 4 y = -3 Green: x = -9 y = -1 |
Displacement: Red: x = 7 y = 2 Green: x = 2 y = 2 |
Multiscale Alignment
For the high resolution images, a multiscale alignment procedure was followed. The images were filtered using a Gaussian Kernel and then scaled down by a factor of half.
An image pyramid is created and the alignment vector of the coarser level images were used as the initial point around which the search window is centred in the
higher resolution levels of the pyramid.
Image: '01657u.jpg' |
|
Displacement: Red: x = 38 y = 0 Green: x = 80 y = 0 |
Image: '01047u.jpg' |
|
Displacement: Red: x = 13 y =-14 Green: x =48 y = 33 |
Image: '01043u.jpg' |
|
Displacement: Red: x = 33 y = -7 Green: x = 107 y =-16 |
Image: '00911u.jpg' |
|
Displacement: Red: x = -54 y = 7 Green: x = 49 y = 42 |
Image: '00458u.jpg' |
|
Displacement: Red: x = 28 y = -26 Green: x = -35 y = -6 |
Image: '01861a.jpg' |
|
Displacement: Red: x = 38 y =-5 Green: x = 80 y = 0 |