Project 3 - Austin Leung

Part 1: Frequency Domain

Part 1.1: Warmup

In this warmup, I simply added the high frequency to the image. The high frequency was obtained by subtracting the original image from a low frequency image.

Original Image

Sharpened Image

Part 1.2: Hybrid Images

In this part, I blended a low frequency version of one image with a high frequency version of another image.

A) Original Images: Adi and Zoolander

Here, I blend a picture of my friend with that of Derek Zoolander. You can see him clearly from close up, but you see Zoolander from far away.

Hybrid Image and Fourier Analysis

Color Hybrid Image

B) Original Images: Campanile and Monet Painting

Here, I blend a photo I took of the Campanile with a Monet painting. It doesn't blend perfectly well, but you can see part of the Campanile and Evans from close up, and the Monet painting from far away.

Hybrid Image and Fourier Analysis

Color Hybrid Image

C) Original Images: California in 2011 and 2014

Here, I blend a picture of California in 2011 and 2014. Unfortunately the pictures were taken from different angles so the landscape doesn't quite line up. Nevertheless, this blending really doesn't work, as you can only see the 2014 waterline.

Hybrid Image and Fourier Analysis

Color Hybrid Image

D) Original Images: Napoleon and a Tiger

Here, I blend a picture of Napoleon and a tiger. This doesn't work perfectly, as you can see both Napoleon and the Tiger pretty clearly from close up. From far away, however, the tiger disappears.

Hybrid Image and Fourier Analysis

Color Hybrid Image

Part 1.3: Gaussian and Laplacian Stacks

In this section, I created a Guassian stack by repeatedly creating low frequency images. The Laplacian stack was created by subtracting the original image recursively from the low frequency images.

Guassian Stack

Laplacian Stack

Part 1.4: Multiresolution Blending

In this section, I blended two images based on a vertical, horizontal, or irregular mask. The laplacian stacks of the images were multiplied to the guassian stack of the mask, then added together to get the blending effect.

Horizontal Blending

Vertical Blending

Irregular Mask

Part 2: Gradient Domain Fusion

Part 2 Project Description

In the second part of this project, the primary goal of was to seamlessly blend an object or texture from a source image into a target image. We begin with a toy problem, in which we reconstruct an image from its gradients. Then, we use Poisson Blending to seamlessly blend a source image into a target image.

Part 2.1: Toy Problem

Original Toy Image

Reconstructed Toy Image

Part 2.2: Poisson Blending

Three examples of Poisson Image Blending

A) Penguin in the Desert

Original images and mask

Poisson Image Blending

A) Penguin on the Beach

Original images and mask

Poisson Image Blending

C) Michael Jordan and a Basketball

Original images and mask

Poisson Image Blending

Laplacian Pyramid Blending vs Poisson Image Blending

In this case, there are pros and cons with both. The Laplacian smooths the blending better, but changes the original image. The Poisson Image keeps the original image mostly the same. The Laplacian Pyramid Blending is likely better for cases where we want to combine two images - like the oraple, while the Poisson Image Blending is likely better for adding an image onto another and smoothing.

Original images and mask

Laplacian Pyramid Blending

Poisson Image Blending