Frequencies and Gradients

Warmup

Original image on the left and sharpened image on the right.

Hybrid Images and Fourier Analysis

Derek and Nutmeg Hybrid

Nic Cage Series

Original images.

Nic and Wolf Hybrid

Nic and Wolf Hybrid Fourier

Nic Fourier.

Nic Low Pass Fourier.

Wolf Fourier.

Wolf High Pass Fourier.

Hybrid Fourier.

Nic and Owl

Nic and Tiger

This one was a failure because the whites of the tiger's coloring overpowered Nic's facial coloring, making it very difficult to make out his face even from a distance.

Gaussian and Laplacian Stacks

Dali

Nic and Wolf

Multiresolution Blending

Oraple

Apple Stacks

Gaussian stacks.

Laplacian stacks.

Mask

Gaussian stacks.

Orange Stacks

Gaussian stacks.

Laplacian stacks.

Moon and Sun

I didn't get the sizes to line up correctly, so ignore that.

Irregular Mask: Moon Crater on the Sun

For the irregular mask, I used the same images as above, but instead of taking half of the sun and half of the moon, I created a circular mask and only used the large crater in the moon.

Gradient Domain Fusion

Brief Overview

The objective of this is to smoothly blend a section of the source image into a section of the target image, using gradients.

Toy Problem Output

Favorite Poisson Blending: Nic and Bunny

Source image.

Target image.

Naive blend.

Poisson blend.

More Poisson Blending: Nic in the Desert

More Poisson Blending: Nic and Mona Lisa

This was a failure. Nic's face is the wrong size and should be rotated, but these are easy fixes. If you zoom in, you can see that the texture of his face is different from the texture of Mona Lisa's face. Since she's a painting, her face has a coarse, grainy texture, which could not be translated to Nic's face. A possible explanation for this is that we only take into account neighboring pixels when calculating gradients. However, since the painting is so coarse, the grains go across multiple layers of pixels, so the texture is not preserved.

Laplace vs. Poisson: Craters on the Sun

Here are the original images.

A crater from the moon is blended onto the sun's surface. It is in the upper right.
The image below was done with Poisson blending.

The image below was done with Laplacian stacks.

In this case, both images turned out well, but the Laplacian one is slightly better (if we do not consider color). If you look closely, you can see the ridges and dimples of the moon around the edges of the crater. These ridges are not visible in the Poisson blended version. This is because Laplacian blends different frequencies and is therefore able to have a smooth transition from sun frequency (smoother) to moon frequency (more dimpled). Poisson focuses more on gradients, so this frequency information is lost. The advantage of Poisson blending is that it preserves color gradients. In the Poisson image, the crater's color gradients are varied and match with the gradients of the rest of the sun. If the Laplacian was done in color, the area of the crater would be discolored because of the original gray moon image.