CS129 / Final Project / Image Hybridization
The Basic Concept
Image hybridization involves changing one’s interpretation of an image with respect to ones distance from it.
At a large distance (or with a smaller image) only the low frequencies are clearly visible.
However up close the high frequencies become apparent.
Einstein Monroe - The Archetypical Example
Method 1: With Pyramids
Generate a pyramid of Gaussians of one image, and a pyramid of Laplacians of the other.
We then choose the first (for one image) or last (for the other) N levels of the Laplacian pyramid to sum. This defines the cutoff frequency.
This creates a pyramid from the top of one pyramid and the bottom of the other.
My Method: In Fourier Space
In taking the Fourier transform of the image, the low frequencies are concentrated in the center of the image and the higher ones in the edges.
If low frequencies are desired, then we mask the outer edges.
If higher, then mask the inner.
Algorithm
- Take the FFT of both images, and use fftshift to rectify them.
- Define a radius around the center of the image, designating the cutoff frequency.
- Loop over the dimensions of the image
- --If the current pixel is within the dimensions of the cutoff frequency radius, include those pixels from the image with low frequency content.
- --Else include the high frequency content.
- Call fftshift again, and take the inverse FFT of the final image
- For color images, repeat the above steps per color channel
Pipeline
Input Images
Fourier Transforms
Masked Fourier Transforms
High Frequencies
Low Frequencies
And finally their Union
More Results
Mike and Rosie
Motor(bike)
And Color too! The Phoenix Suns
