Images of the Russian Empire:
Colorizing the Prokudin-Gorskii photo collection
15-463: Computational Photography
Ming Han Teh
Background
The requirements for this project can be found on the course website.
Technique
The image strip is composed of BGR glass plates.
I divided the image strip into 3 equal parts (along the height).
I then aligned the image by using exhastive search with an image pyramid of scaling factor 0.5.
For this pyramid, my top most image must have a width of less than or equal to 300 (i.e. the base case).
For the exhastive search, I used a window of [-15,15] for my top most image and [-2,-2] for the rest of the images.
I tried both the Sum of Squared Difference (SSD) and Normalized Cross-Correlation (NCC) to compute the metric,
but did not found much difference in results in the two.
Hence, I settled on SSD as it was fast and more efficient.
This technique worked well for some small images and the alignments were still off for most of the images.
In implementing the bell and whistles, I realized that the white and black borders were affecting the metric scoring and hence the alignment was off.
I then used a navie algorithm to crop away the white and black borders on the original image strip before processing it.
This algorithm was implemented by taking means along each columns and rows, and then thresholding for black and white via a for-loop along both axis.
I also used used the edges() function for the top image on the pyramid before using the search + SSD algorithm on it.
This was used for the smallest image to match the key features of the image.
However, it was not used for the rest of the images because it resulted in too much imprecision.
Finally, I also implemented the autocrop for the final image to trim away some of the color bands at the edges.
This was implemented by calculating the SSD for each pixel for each R,G,B pairs and suming up.
I then calculated the mean along the columns and rows, and threshold them via a for-loop along both axis.
Images
00088v.jpg
Result 1:
displacement of [-3,0] of Green wrt Blue
displacement of [5,-1] of Red wrt Blue
Result 2:
displacement of [-9,3] of Green wrt Blue
displacement of [-19,5] of Red wrt Blue
01880v.jpg
Result 1:
displacement of [6,2] of Green wrt Blue
displacement of [12,2] of Red wrt Blue
Result 2:
displacement of [-14,2] of Green wrt Blue
displacement of [-26,4] of Red wrt Blue
00757v.jpg
Result 1:
displacement of [0,0] of Green wrt Blue
displacement of [4,0] of Red wrt Blue
Result 2:
displacement of [-8,3] of Green wrt Blue
displacement of [-14,5] of Red wrt Blue
01031v.jpg
Result 1:
displacement of [1,1] of Green wrt Blue
displacement of [4,1] of Red wrt Blue
Result 2:
displacement of [-5,1] of Green wrt Blue
displacement of [-8,1] of Red wrt Blue
00911v.jpg
Result 1:
displacement of [2,-1] of Green wrt Blue
displacement of [6,-3] of Red wrt Blue
Result 2:
displacement of [-10,-1] of Green wrt Blue
displacement of [-10,-2] of Red wrt Blue
00888v.jpg
Result 1:
displacement of [6,0] of Green wrt Blue
displacement of [10,-1] of Red wrt Blue
Result 2:
displacement of [-7,1] of Green wrt Blue
displacement of [-14,0] of Red wrt Blue
00889v.jpg
Result 1:
displacement of [1,-1] of Green wrt Blue
displacement of [5,-1] of Red wrt Blue
Result 2:
displacement of [-7,2] of Green wrt Blue
displacement of [-13,3] of Red wrt Blue
00106v.jpg
Result 1:
displacement of [4,0] of Green wrt Blue
displacement of [9,-1] of Red wrt Blue
Result 2:
displacement of [-4,1] of Green wrt Blue
displacement of [-6,-1] of Red wrt Blue
00137v.jpg
Result 1:
displacement of [6,4] of Green wrt Blue
displacement of [10,7] of Red wrt Blue
Result 2:
displacement of [-7,5] of Green wrt Blue
displacement of [-14,8] of Red wrt Blue
00907v.jpg
Result 1:
displacement of [0,-1] of Green wrt Blue
displacement of [4,-1] of Red wrt Blue
Result 2:
displacement of [-7,0] of Green wrt Blue
displacement of [-12,-1] of Red wrt Blue
00029u.tif
Result 1:
displacement of [77,-6] of Green wrt Blue
displacement of [88,-12] of Red wrt Blue
Result 2:
displacement of [-84,15] of Green wrt Blue
displacement of [-156,32] of Red wrt Blue
00087u.tif
Result 1:
displacement of [54,0] of Green wrt Blue
displacement of [102,1] of Red wrt Blue
Result 2:
displacement of [-75,38] of Green wrt Blue
displacement of [-138,55] of Red wrt Blue
00128u.tif
Result 1:
displacement of [36,2] of Green wrt Blue
displacement of [85,3] of Red wrt Blue
Result 2:
displacement of [-41,25] of Green wrt Blue
displacement of [-97,38] of Red wrt Blue
00737u.tif
Result 1:
displacement of [38,-2] of Green wrt Blue
displacement of [49,-1] of Red wrt Blue
Result 2:
displacement of [-81,6] of Green wrt Blue
displacement of [-142,14] of Red wrt Blue
00822u.tif
Result 1:
displacement of [62,3] of Green wrt Blue
displacement of [101,-4] of Red wrt Blue
Result 2:
displacement of [-65,24] of Green wrt Blue
displacement of [-118,33] of Red wrt Blue
00892u.tif
Result 1:
displacement of [-16,-8] of Green wrt Blue
displacement of [59,-15] of Red wrt Blue
Result 2:
displacement of [-59,1] of Green wrt Blue
displacement of [-109,4] of Red wrt Blue
00992u.tif
Result 1:
displacement of [49,-5] of Green wrt Blue
displacement of [73,-13] of Red wrt Blue
Result 2:
displacement of [-76,14] of Green wrt Blue
displacement of [-138,20] of Red wrt Blue
01043u.tif
Result 1:
displacement of [-18,3] of Green wrt Blue
displacement of [48,-11] of Red wrt Blue
Result 2:
displacement of [-86,10] of Green wrt Blue
displacement of [-126,16] of Red wrt Blue
01085u.tif
Result 1:
displacement of [97,10] of Green wrt Blue
displacement of [104,23] of Red wrt Blue
Result 2:
displacement of [-71,31] of Green wrt Blue
displacement of [-117,59] of Red wrt Blue
01734u.tif
Result 1:
displacement of [8,-2] of Green wrt Blue
displacement of [212,-4] of Red wrt Blue
Result 2:
displacement of [-34,28] of Green wrt Blue
displacement of [-61,49] of Red wrt Blue
00895v.jpg
Remarks: Example of own choosing from LoC website.
Result 1:
displacement of [7,2] of Green wrt Blue
displacement of [12,0] of Red wrt Blue
Result 2:
displacement of [-8,2] of Green wrt Blue
displacement of [-14,3] of Red wrt Blue
00299v.jpg
Remarks: Example of own choosing from LoC website.
Result 1:
displacement of [1,0] of Green wrt Blue
displacement of [7,0] of Red wrt Blue
Result 2:
displacement of [-11,0] of Green wrt Blue
displacement of [-12,2] of Red wrt Blue
00645u.tif
Remarks: Example of own choosing from LoC website.
Result 1:
displacement of [28,-9] of Green wrt Blue
displacement of [103,-17] of Red wrt Blue
Result 2:
displacement of [-94,6] of Green wrt Blue
displacement of [-193,-9] of Red wrt Blue
00234u.tif
Remarks: Example of own choosing from LoC website.
Result 1:
displacement of [40,-8] of Green wrt Blue
displacement of [138,-21] of Red wrt Blue
Result 2:
displacement of [-101,5] of Green wrt Blue
displacement of [-143,13] of Red wrt Blue
Hybrid Images
Background
The requirements for this project can be found on the course website.
Technique
I use the guassian filter for the low pass filter, and for the high pass filtered image, I substracted the gaussian-filtered image from the original.
Images
DerekPicture.jpg vs nutmeg.jpg
algore.jpg vs jobs.jpg