Project 1

Juefei Xu

juefeix@andrew.cmu.edu

Sept.9 2010

Methodology:

(1) For low resolution images, exhaustive search is applied on window of [-15,15]. Because each channel in low resolution image is about the size of 380x320, so a window of [-15,15] is quite enough for aligning those channels.

(2) For high resolution images, whose size is about 10 times the size of low resolution image, exhaustive search would be too slow because it requires much larger window for aligning. If the window size is 10 times larger, say [-150,150], the search would be impossible to finish in a few minutes. So a image pyramid is used to speed up the search. In my implementation, I started builing the pyramid from image size of 256x256 to 4096x4096 which exceeds the size of the high resolution image. Do exhaustive search in the 256x256 image first, align the image, upgrade to a larger size and than do exhaustive search again. In that way, the window size for displacement can remain small. In my case, I choose it to be [-35,35]. The reason I start from 256x256 instead of 4x4 is that 256x256 with window size[-35,35] is already very fast to implement.

The high resolution images are cropped to proper display on the website. The original size image can be generated from my code. The results are shown below:

I.Low Resolution Images:

(1)00125v