Proposal

For the final project, I am going to implement the paper by L. Zelnik-Manor and P. Perona, "Automating Joiners." Joiners is a term used by artist David Hockney to describe an art of putting many pictures together to create a mosaic. However, this mosaic is different from panoramas we usually see because the photographs used to produce the mosaic need not be taken from the same view point. The pictures also need not be coherent with each other. These differences create a distinct effect for joiners to capture something that a regular panorama cannot capture.


An example of a joiner by Midwest Mike

The paper introduces a way to automate the process of joining different photos in a set together. It can be broken into 4 main steps

1. Find feature correspondences between each pair of pictures. The paper uses the Lowe's SIFT features. The site linked also provided MATLAB code which I intend to incorporate in the project.

2. Find the best alignment for all the picture matched.

3. Find the best order for the alignment.

4. Using the knowledge from the alignment in 3, realign the image again to disregard the covered features and give more weights to the features near the boundaries.

5. (Optional) User gives input to make the images aligns better or look more artistic. User input can also be used between step 1 and step 2 to match rejected photos with each other.

In addition to what is described in the paper, I will also explore other blending types. An example of that would be to blend the mosaic using user defined masks to help protect some important part of the picture.

If time permits, step 5 will be implemented with a GUI.

Back to final report