Gradient Domain Fusion

Shiva Peri - Assignment 2

DESCRIPTION

The following images employ the use of the Poisson image blending. The technique arises from the insight that gradients matching takes priority over intensity matching. In other words, the shape of the image matters more than the colors. In Poisson Image Blending, the gradients are formed into a least-squares problem using a sparse matrix. The resulting image is then re-shaped into the final output dimensions. Each color channel is blended individually and the blended channels are recomposited at the end

FAVORITE RESULT

source image rescaled source image source mask unblended masked image blended masked image

OTHER RESULTS

FAILURE CASE: This case fails to preserve the white upper part of the cow. The source image had a white sky and a green ground. Because of how the boundary pixels are handled by the blending algorithm, this results in a green gradient across the cow.
Here we have a shark fin in the water. This result blends well, due to the similarity of the background colors of the source and target images.
The same target image does not work as well for this ship, due to the differences in the background of the source image.