Matt Ricketson
Computational Photography
Assignment 3: Homography

The images below show successful tests of image homography. The posters were transformed to fit into office picture in a natural wall, conforming to the surface they are supposed to be on top of. The goal is to try and add an object or feature into a scene synthetically while stil maintaining realism.

Here are the original pictures, including two posters and an office scene:

Poster 1 Poster 2 Office

Here are results from using homography to insert the posters into the image:

Poster 1 Homography Poster 2 Homography

There are several important things to note about this project.

First, this implementation is not user-proof. I found that if I did not select corresponding image points in a proper clockwise fashion, the resulting homography would not look correct.

Second, even though this implementation allows a feature to be inserted, it does nothing to guarantee that it is inserted correctly. For instance, it is up to the user to specify a destination region that aligns with the natural perspective view in the image. To correct this, additions could be made to the algorithm to try and self-correct target points or to select the target points without any user intervention required int he first place. This would require a good deal more to be added to the algorithm, however.

Finally, a difficult aspect of the project involved making sure that the transformed poster was correctly inserted into the scene. This involved a double-mask method, with one mask cutting out the destination region from the office image and another mask cutting out the transformed poster to the correct dimensions. After both masking jobs the results are added together to create a seamless image. To make sure that the transformed poster was positioned correctly it had to be translated onto an image of equal dimension to the office picture, and then masked.