Goals

Work In-Progress and Future Goals

The following is a list of things that I am either still working on or would like to add in the very near future.

Multi-size Video Tiles
Support for this is a work in progress. I would like to figure out a way to have a tile expand its size if the search determines a better match can be found with a larger tile. Some architecture changes will be required before this can happen, but I really think it will make the results more interesting.

Spatial Hashing
Currently, the video mosaic search is a linear one. I have heard of hash functions that effectively partition high-dimensional space into buckets. If I could implement something like this into my search, I anticipate orders of magnitude of speed-up, which in turn would allow me to consider many, many more tiles.

Experiment with Feature Weighting
All the machinery is in place now for proportional weighting of features. Extremely limited experiments show that even when normalized, some features completely dominate others. It would be great to build a matrix of results for each feature so as to do a quantitative analysis.

Find Constants for Tile Movement
Currently tiles can move independently, but I have yet to find a good quantitative way to have them move at the correct multiple of the flow value. This will take experimentation.

Region Recognition for Bigger Tile Mosaics
With the metrics I have, mosaics with smaller tiles look decent. However, once the tile size increases, the metrics break down and visual result quality decreases. I been trying to think of ways to solve this problem and would like to try some sort of region isolation--i.e. perhaps edge detection and then filling in the resulting image? Finding color blocks? Something of that nature might be useful here.

Simple Color Correction
Although I viewed this as 'cheating' before, I now consider it as smoothing a noisy signal. I believe remedying the greatest differences in color between movies would make a large difference.

Sections