Computational Photography Project 1

Nick Vandal

Overview

This project explores gradient-domain processing, a simple technique with a broad set of applications including blending, tone-mapping, and non-photorealistic rendering. For the core project, we will focus on "Poisson blending" described in Perez et al. 2003 paper

Toy Problem Reconstruction

Compute the x and y gradients from an image s, then use all the gradients, plus one pixel intensity, to reconstruct an image v.

Poisson Blending

Solve the blending constraints:

Mixed Blending

Solve the blending constraints:

Blending Examples

Source Image Target Image Direct Pixel Insertion Poisson Blending Mixed Blending
Poisson blending works fairly well...the blur around the edges of the shuttle looks like heat during reentry. Mixed blending allows the mountains to be visible from underneath the shuttle.
Poisson blending works okay, but leave Colbert excessively orange.
Poisson blending is works fairly well. Mixed blending is too transparent.
The astronaut blends well with either technique.
The jet has superior blending performance with Mixed blending.
The penguin has better blending using Mixed blending