Project 2 Writeup

Soumya Ghosh (sghosh)
January 29, 2010

This project involved blending images using the Discrete Poisson Solver suggested in Perez'03. My implementation here follows the paper pretty closely. It solves a system of linear equations, to figure out the appropriate values of the blended image pixels. The system of linear equations was expressed in the matrix form as suggested in the handout and was solved using a QR decomposition of the constraint matrix. I used speye command to create the sparse matrix, which saved me a significant amount of time by not having to access the vast majority of enteries in the sparse constraint matrix which do not lie under the mask. My implementation doesnot do anything clever with the pixels under the mask however. This causes the processing to be slow when large masks are involved.

Extra Credit: For extra credit I implemented Burt and Adleson's multi-resolution image blending scheme. The results for most test images are somewhat worse than the Poisson Blending scheme. The boundary regions, under this scheme, appear more smoothed with distinct artifacts visible. However, this method appears to be significantly faster than the poisson blending method.