Blending represents fusing of two or more images to form a single composite image. Normal cut and paste termed as naive blending results in non smooth transition between the boundaries of images. Image blending in gradient domain simplifies to a Poisson partial differential equation. This project explored the effects of Poisson blending and it’s modified version (obtained by slightly modifying the constraint) on blended image quality. Since Poisson blending typically involves inversion of a sparse matrix, project also investigated performance of various sparse matrix inversion methods like Gauss-Seidel and Sparse LU Decomposition on quality of blended image.
This project was undertaken as part of Advanced Image Processing (E9 246) course during Jan-May 2017. Further details about the project and results are available here.