David Dominguez Hooper

Background

Sergei Mikhailovich Prokudin-Gorskii (1863-1944) photographed thousands of color pictures by recording three exposures of every scene onto a glass plate using a red, a green, and a blue filter. For the purpose of this project, I used digitized Prokudin-Gorskii glass plate images to generate clear color images by extracting each color channel and aligning them on top of each other.

Algorithm

I first tried the simple naive approach of using two for loops to go through the small .jpg images by rolling the image by an offset of [-15 to 15] pixels in both the x and y directions. I tried both SSD and NCC to calculate the best displacement for each of G and R channels that match the B channels the best. I found that SSD gives better runtimes for processing images, so I stuck to that. This worked for the small .jpg files, but took too long for the larger .tif files. So, I used the pyramid approach to scale down the large .tif images to 1/16 scale to start off. I then used SSD as before on the 1/16th scale image to get an intital displacement. After I get the displacement, I increase the scale by a factor of two and repeat the process until we are back at the full image size. An important note is that the next displacement values for x,y are a function of the previous displacement values, i.e. x_new = (x_old)*2 + x_dis, y_new = (y_old)*2 + y_dis. I also pre-process the image by cropping 10% of the border in order to give better results when calculating alignment.

Challenges

The first challenge I faced was when I approached the pyramid algorithm. I tried a recursive implementation but could not get a working solution. So, I tried an iterative approach and had success with that. The second challenge I had with alignment was with the Emir photo. I could not get good alignment using the base image as the blue channel. I tried using the green channel as the base and found the displacements for the blue and red channels and had success.