Assignment #2 - Poisson Blending

Rohan Choudhury

I. Project Overview

Overview

In this project, we implemented poisson blending, which aims to place regions from a source image into a target image. To make it more natural, we set up the blending as a least squares problem, so that the image gradients in the new target image are taken from the source. This results in a more natrual blend, but is obviously not foolproof, as shown in our results below.

II. Results

Toy Reconstruction

The result of the toy reconstruction.

Poisson Blended Images

Jaws in a Swimming Pool
Penguins On the Ski Slope
Sharks In A Swim Race

Of the above images, the swimmer and the sharks is my favorite. It turned out the smoothest. The worst result was the jaws shark in the pool. I believe this is because there is a large difference in the image content, intensity ranges, and colors in the region that is above the pool, and the region below the pool is a more uniform blue. This results in an odd looking blend. This implies to me that Poisson blending works well in certain situations (such as when the target region has similar colors), but works poorly if the target region crosses multiple color regions or is entirely different.