CSCI 1430 Project 1
Hybrid Images
Reese Kuppig (rkuppig)
The objective of this project was
to implement an algorithm that combines two images into one
hybrid image using Gaussian and Laplacian
image pyramids. The overall effect is created by separating the high
and low frequencies of the images such that one image is more visible
from close-up, while the other image becomes more visible as you move
farther away.
Image
Pyramids:
Guassian and Laplacian pyramids represent a repeated low
pass filtering and high pass filtering, respectively, of an image. Each
pyramid level is formed as follows:
Results:
1)
The original image is blurred with a Gaussian filter, and the blurred image is
subtracted from the original, essentially extracting the highest
frequencies in the image.
2) The
blurred image is then downsampled, to negate the effects of
the Gaussian filter, and step 1 is repeated for this new
image.
Below is a sample Gaussian image pyramid,
followed by a Laplacian image pyramid. The Laplacian pyramid has been
brightened for clarity.
Algorithm:
The process of creating a hybrid image breaks down
into the following steps:
1) The two images
are aligned such that similar features overlap
and mask eachother. This aids in the visual effect, so that noticeable
features of the less visible image do not distract the viewer's perception of
the dominant image at a certain viewing
distance.
2) Each image is
then decomposed into a Gaussian pyramid and a Laplacian pyramid
as described above.
3) Given a cutoff index N, the hybrid
image is composed by combining the first 1 through N levels of the first
image's Laplacian pyramid with the N+1 through last levels of the second image's Laplacian
pyramid and the last level of the second image's Gaussian pyramid.
4)
Finally, the hybrid image is cropped and exported.
Results:
The two
original images followed by large and small versions of the
hybrid image. Notice that the first image (high frequencies) is more
prominent in the large hybrid image, while the second (low frequencies)
becomes visible in the small hybrid image.
Some image pairs
are better suited to the effect than others.
For instance, the Einstein-Marilyn hybrid works well, probably due to
the similarity of tone and shape of their faces,
whereas the rhino-car hybrid does not work quite as well, possibly due
to the prominent areas of solid color in both images, which are different shapes
and do not overlap well in the composition.
 
 
 
 
|