6.815/6.865  Adv Computational Photography

Reminder information on class project presentations (for Thursday's class):

2 minutes of presentation time per student (so 2-person teams get 4 minutes, 3-person teams get 6 minutes).
You only have enough time to state the problem, say a little bit about what you did, and show your results.
Please e-mail your Powerpoint or pdf slides to Berkin (berkin@mit.edu) by 10am Thursday morning.

Good luck finishing up the projects, and I look forward to hearing about them on Thursday.  I'll bring juice and cookies...

Car keys were found after class today.  Please come by my office 32-D476 or call me 617 253-8828   to arrange to pick them up.

--Bill

Thank you, if you have filled-out the class evaluation form.  If you
haven't yet, please do.  We read them carefully and it gives us valuable
feedback for improving things next time.

https://sixweb.mit.edu/student/evaluate/6.815-s2010
https://sixweb.mit.edu/student/evaluate/6.865-s2010

Thanks,

Bill

Hi,

Please take a moment to fill the course evaluation form and grade us:
https://sixweb.mit.edu/student/evaluate/6.815-s2010
https://sixweb.mit.edu/student/evaluate/6.865-s2010
Thanks,
-berkin

Even though we didn't cover it in class, the problem set asks us to normalize the feature points. This way we should be able to use the same threshold when determining inliers for any size of input image. Let us try to see how we can translate the mean of the points to the origin:

[p';1] = [1 0 -mean_x; 0 1 -mean_y; 0 0 1] * [p;1]
where mean_x and mean_y are the average coordinates of the selected features. Call this translation matrix Trans.
Now we need to scale the vectors' norms:

[p'';1] = [s 0 0; 0 s 0; 0 0 1] * [p';1]
where  s = sqrt(2) / average_L2_norm  is the scaling factor. Call this scaling matrix Scale. Then,
T = Scale*Trans  should give us the transformation matrix we are looking for.

-berkin