For this project, I implemented laplacian pyramids to create an effect known as hybrid images. The general idea is that we want to create an image ythat looks differently depending on how it is seen. If it is seen from a short distance, you will see one image, and if at a distance, you will see another. This is done by splitting the images into what are called low and high frequencies. The high frequencies represent the sharp lines and fine detail of an image, while the low frequencies contain the general shape and intensity without much detail. The high frequencies are visible close up and dissapear as the image gets smaller/farther away, while the low frequencies do the reverse. Thus by superimposing the high frequencies of one image over the low frequencies of another, we will see the the high frequency image when close up and the low frequency one from far away.

As menitoned before, these hybrid images are created by using a laplacian pyramid. The general idea is that we progressively make the picture blurrier and blurrier, and extract out the difference of the blurs . We start with the base image and apply a gaussian blur. The first image of the pyramid is the difference between the blurred and original image. We then shrink our blurred image by a factor of 2 and repeat for some number of iterations. The earlier images represent the high frequencies because they have the fine detail which is lost by blurring. The latter images constituate the low frequencies because after numerous blurs, all of the fine detail will be gone. To determine just how many of the early and later images of each pyrmaid to use, I just manually tried numerous cutoffs and stuck with the one that I judged to be the most aesthetically correct.

In some cases where the images are very similar to one another, this technique works very well. For example, making a Cat-Dog hybrid works reasonably well.

In addition, adding human faces works relatively well. Here we have included an example of a hybrid between Jeremy Clarkson and our own Tom Doeppner, as well as the classic mixture of Einstein and Marilyn Monroe.

Unfortunately, things that don't already have a high level of similarity don't work quite as well. Combining Arnold Schwartzenator with his terminator form fails to please the senses because the robot form will always clache with human tones.

Next, we try combining a face with a completely random object, specifically Carl Sagan and the Milky Way

The high frequencies of Sagans face completely clache with the universes low frequencies, making for an odd image when seen close up. From afar it works better as Sagan's high frequencies are lost from vision, but some odd jagged lines persist.

For our final task, we try combining 2 objects that are not faces and see what happens.

Up close the car is fairly detailed, but the lighting looks very odd due to the rhino low frequencies. Like before though the low frequency image looks fairly decent with just a few odd lines that are easily ignored on a quick inspection. Only afteer a bit of study can you uncover the other high frequency image, and even then, only a little.

As a final twist, we try using the scheme but with color. The algorithm was the same as before, except now we don't grayscale the images first. Our "best" result came from combining Jeremy Clarkson and Tom Doeppner.

Aside from some discoloration, the affect works pretty well. With this "success" we might expect catdog to work well in color.

The low frequency dog looks pretty good, and the outline for the high frequency cat works fairly well. However we completely lost the color of the cat, which is expected since it is a very low frequency piece of information. For the finale in our face series we try the Governator in color

Once again the low frequencies look good, but the large low frequency image works very poorly. While the outline of his portrait is visible, the color for the ternimator just makes no sense and makes the image look odd.

Now we try hybrid color iamges for things that are not similar. First we see the Sagan Galaxy

Just like before, the low frequency galaxy looks good, but the high frequency sagan portrait does not blend with the colors of the milky way. Next up is the rhino car.

Once again the rhino colors completely clash with the car and make it visible but inconcievable. In general it seems like the method can work to make decent hybrid color images, but the images being combined must have similar color pallets or else the effect will just be uncanny. Like the X-Men.