Instructions for GCV-based image deblurring.

1.
Download from web site
 deblur1_data.mat
 deblur1_gcv_template.m
 deblur1_rms.m

2.
Run deblur1_rms.m (using toolbox)
to see how NRMSE varies with regularization parameter
for a simple example.
Ask questions about any parts of this example that are unclear to you!

3.
Run deblur1_gcv_template.m
This template uses the wrong formula for GCV
so it picks the wrong value of the regularization parameter.

4.
Make a copy of the template, e.g., deblur1_gcv_myname.m
Modify it so that it computes GCV correctly (for a nonlinear estimator).
You will need to adjust the search range for beta to minimize GCV properly.

5.
Enter the optimal value of log2(beta) that you found into the google doc
for class 8 under "homework" on web site.

6.
If time permits, use GCV to experiment with delta and/or other parameters.

7.
If time permits, use GCV based on the linear circulant model to select \beta
and see how close it is to the one based on "nonlinear GCV" above.

If you do 6 or 7, then email me some code/results/comments.
Please do not send me .fig files (useless in octave).  Send pdf or jpg.