Matthew Banner

15-463 Fall 2007

The images below were created using the following algorithm:

Split the original image into three equal-sized pieces with each piece corresponding to one color layer.
Create an image pyramid for each layer. Because image intensity will naturally vary between the different layers, it was desirable to find a means of aligning based on the features (e.g. edges) of each layer. To this end, rather than storing different sized copies of the original image in the image pyramid, each size of the original image was run through a high-pass filter and the result was stored in the pyramid.
Using the red layer as a baseline, find the offset of the other layers by stepping through the image pyramids from the smallest to the largest and comparing various alignments using the Sum of Squared Differences distance.
Using these offsets, align the layers and crop out the portions of the image which are outside the bounds of any of the three layers.