CS 143 - Project 1: Hybrid Images with Laplacian Pyramids

Arman Uguray (auguray)

In this project, I combined the high and low frequencies from two different images to create a hybrid image. The seperation of frequencies is done by constructing a Gaussian and a Laplacian pyramid for each image. Then, a cutoff frequency is selected and a new Laplacian pyramid is constructed by combining the lower frequencies that are above this threshold from the pyramid that belongs to the image which was selected for the lower frequencies and the higher frequencies from the other pyramid. Then the hybrid is constructed by taking the top level image in the gaussian pyramid of the image that was selected for the lower frequencies and additively iterating through the newly constructed laplacian pyramid.

Here are some examples:

1. Humanzee

The first "Humanzee" hybrid takes its low frequencies from the picture of the chimpanzee and the high frequencies from the man. The dark colors of the chimp are visible in the larger image, however the man's features are the most prominent (especially when seen from very close).

humanzee humanzee

The second example takes the low frequencies from the human and the high frequencies from the chimp. Even though the larger image looks quite convincing, the eyes of the chimp are still apparent in the smaller image. One needs to really step back from the screen to appreciate the smaller image.

humanzee2 humanzee2

Here is a different version of the second example, with a different alignment. The hybrid looks less convincing in this one, and the scaled-down version is not as good as the previous, since the chimp's eyes are still visible on the man's forehead. This is a good example of the importance of good alignment of eyes in facial hybrids.

humanzee2 humanzee2

Originals:

chimp guy

2. Mona Bean

This hybrid is a combination of Da Vinci's Mona Lisa and Rowan Atkinson.

connery connery

Originals:

old-connery young-connery

3. Sean Connery

This hybrid combines a picture of an elderly Sean Connery with one from his youth. The alignment in this image is particularly good, however since Sir Connery's elderly moustache is not completely filtered out by the Laplacian filters, it is still visible in the smaller image, which makes him still look old.

connery connery

Here is another hybrid with a slightly adjusted alignment and a different cutoff. Notice that the previous problem has been eliminated, but the older features are not as pronounced in the larger picture. The desired effect can be achieved by further scaling up the larger image.

connery connery

Originals:

old-connery young-connery

Here are some other examples:

cat-dog einstein-monroe derek-nutmeg