I doubt anyone noticed, but I was absent a lot this fall. Whole days would go by without a visit to MR. I was gone because I was taking several classes.
I just finished a project for one class, and I'm proud enough that I wanted to share it with you. It's pretty technical, but I bet there are a number of people here who will understand it.
I wrote a program that matched a picture downloaded from Google images with one found in a target photo. For example, I did a Google search on "spam" and then used one of the results. The goal is to find enough similarity between the 2 images that you can reliably say that they show the same thing.
I'm sure you're thinking that this is easy. Did you know that (I think) a quarter of the human brain is devoted to vision? A significant portion of the vision processing in the human brain occurs without conscious thought. Also, it took your brain years to learn how to see objects. This is not an innate ability (I've worked with teenagers, so I know this for a fact). Duplicating this ability is still an interesting problem in computer vision.
some technical detail
The project was written based on the goals set in
The Semantic Robot Vision Challenge. The program was written in matlab, and I relied heavily on VLFeat library. I used
Harris corners to select the points to try to match, and
SIFT to describe the points and match them.
With some refinement, my program is good enough that I'm planning to use it to compete next year. Some success images follow. Enjoy.