Log Update

Feb 11, 15:00-16:00 - Met with Glen, Jeff and Crystal. Discussed Requirements Document.

Feb 13, 12:45-13:45 - Filled out Scope of the System and Use Case List for Requirements Doc.

Feb 13, 14:00-15:00 - Met with Jeff and Crystal to discuss Requirements Document. Mostly, Q&A for Crystal and Jeff to understand what the project is about.

Image Segmentation Proposal

Proposed image segmentation:

original:


segmented:

Log Update

Feb 04, 15:00-16:00 - Met with Glen, Jeff and Crystal. Discussed initial project direction for Practicum. Jeff and Crystal are going to iron out the requirement docs, while I provide a description.

Feb 05, 19:30-20:00 - Met with Jeff. Worked on description and problem definition.

Coding Started

I've been working on a simple visual output class for debugging and testing. Simply put, it's an SDL and OpenGL wrapper with functions geared towards viewing the work being done. Hopefully it will eventually provide a simple interface for the user to interact with the app.

The code is available on SourceForge.

I'll be working next on providing a short how-to on getting started with the project. This info will be available on the project's wiki.

Need to make sure that this project is adhering to GPL licensing!

Formalizing Documentation

I've started formalizing the Virtual Cam documentation of the research and development. I'm going centralize all work on a wiki page. It will also serve as a date verification and give it a better web presence.

Over the winter break and interterm, I'm going to be working on a base for the project. The base will have a very basic API to load the image sequences and perform all the prepping. It will probably have a rudimentary visual output window so we can see "progress" and debug. More than likely, it will be written using SDL or OpenGL, as they are great cross-platform solutions that don't require much of a learning curve.

I will probably also start start saving the code to the SourceForge site.

Wiki Site: http://virtualcam.wikidot.com

Texture Patches

is there some good way to parametrize a triangular texture in order to compare two textures for similarity?

links to look at:

• graph cut matching in computer vision. (toby collins)
• introduction to min-cut/max-flow algorithms. (hong chen)
• graph cut algorithms in vision, graphics and machine learning. (sudipta n. sinha)
• graph cuts. (many)
  

outline:

orig


dark outlines with shadings


lightened shadings


need a way to follow outlines and build texture swaths. we're not going to track the texture or points in the texture (features) as SIFT might. instead, we're tracking and following what swaths are adjoined... we're going to expand on the unwrap mosaics, allowing for more dynamic scenes and removing the assumption of cylindrical topology. it will only take many, many intermediate steps to get there. the idea thus far will parallelize very easily! might even run decently on a GPU...

using statistics to aid tracking. histograms, gradients, mean/median, standard deviation.

build a "flattened" image of dynamic resolution... that's the "trick!"

2008 Oct 28

I've begun trying to really understand the math behind the Unwrap Mosaics. So far, everything is quite straight-forward. The only really hurdle I'm finding is that I don't have nearly enough math behind me to really grasp what is being said. (approx 1hr)