Programming Project #1
15-463: Computational Photography

Assignment

The results of the assignment were very successful on both low resoluation and high resolution images. I used the normalized cross-correlation function available in Matlab (normxcorr2) to come up with a displacement metric for each image. This was combined in a small recursive function to resize large images into a smaller, more manageable image. The "base case" for the smaller image was found through trial and error at the point where the NCC formula began to break down.

With a computer with 2GB of RAM, I was not able to process the full-size, high resolution TIFs. However, after resizing them to approximately 80% of their original size (the final plate image size was roughly 3000x7800 pixels in each case), I was able to process the images relatively quickly, with the same quality of results as the low resolution JPGs.

The Matlab program consists of two functions:

  • colorize_zmouri(displacement, image_name) - the primary function, this reads in the image name and calls the colorize function. It returns the offset vectors for the G and R images.
  • color_helper(displacement, image) - a recursive function that will crop the image according to the displacement, resize the image if necessary, and compute the offset.
To execute the Matlab program, run colorize_zmouri with the specified image displacement (15 gave sufficient results) and the name of the image. For example,
EDU>> [gOffset, rOffset] = colorize_zmouri(15, '00125v.jpg')

gOffset =

     5    -1


rOffset =

     9     0

Image Results

The following images are the results of combining each of the plate images. Click on each one for the full resolution picture:
Low Resolution JPGs
High Resolution TIFs - warning, each image is ~45MB in size
Additional Images from the Library of Congress

01848 Elderly Sart man (Babaika), Samarkand. (JPG, 27KB)
01882 Kush-medrese (interior from the right side). Bukhara. (JPG, 32KB)
00626 Excavation near Katav-Ivanovskii Zavod. (TIF, 45MB)