CPSC 535 - Computer Vision & Image Analysis   TA PAGE

There will be no lab today.

On Thurdsay, I will do some review for the final exam.

Labs this week will be Q&A style. If you have questions related to your assignments or any lecture material, come and speak to me.

Whew! All assignments that I've received as of today have been marked.

If you have any question about a mark received -- or if you are missing a mark -- please send me an email or find me in lab.

Today I covered Assignment 5 - 3D Vision. The main theory behind it is camera calibration, which was covered this week in lecture.

We also talked about problems relating to A3 and A8.

Today I covered Assignment 7 - Image Registration, mostly how to build the homography.

In today's lab we talked about the Harris corner detector, and covered a numeric example to see how it worked.

The Harris corner detector can be used to automate a portion of the image registration assignment (A7).

There is no lab today due to Reading Days.

Much apologies for not updating this site for awhile. Here's a quick summary of the recent lab topics:

• Oct. 24 This lab was Q&A.
• Oct. 26 Midterm review.
• Oct. 31 Midterm - no lab.
• Nov. 2 Covered some Assignment 3 tips.
• Nov. 7 Linear algebra review, including least squares, eigenstructures, and singular value decomposition (SVD).

Today's lab will be Q&A, since hopefully everyone is working on assignments for next Monday!!!

I haven't received very many assignment submissions so far, as you can see by looking at the grades table below.

This is a friendly reminder that if you want to submit and get a mark for all five assignments, you have to hand in a second assignment by Nov. 13, which is three weeks from today, and one assignment per week after that.

The freedom offered in this course can be a blessing if you use it to avoid the typical 5-course crunches that happen, but it can also be a curse if you leave the assignments to the last moment.

</sermon> ;)

Next week's labs will be Q&A-style. If you have any particular topic you'd like to see covered in lab, please send me an email.

In today's lab I gave an image processing quiz. Answers are on the last page.

There is a great resource [octave.sourceforge.net] for Octave implementations of many useful operations, such as

• Image processing techniques from class, such as morphological operations (see erode, dilate), binary image functions (bwarea, bwlabel), and various filters such as median filters, contrast stretching, and so on.


• Image loading functionality similar to that found in Matlab (imread, imwrite, ...).


• Tons of other stuff that may or may not be relevant to this class.

I would recommend grabbing some of them and trying them out on different images (if the server ever comes back to life).

Here's all the stuff needed to play around with the masks/operations discussed in lab today:

• [histeq.m] Histogram equalization code for Octave. (Based on discrete HE from here.)


• [images.zip] Image archive.


• [ipquiz.pdf] Examples and a short quiz (answers on last page).

Example usage:

[f, t] = pnm_read('simple.pgm');
f_he = histeq(f);
f_box3 = conv2(f, ones(3,3) / 9, 'same');
f_log = conv2(f, [0 -1 0; -1 4 -1; 0 -1 0], 'same');

colormap(gray(256));
image(f); % image(f_he), image(f_log), ...

Histogram equalization (HE) was covered in lab. HE is part of a more general class of point operators in image processing; that is, operators for which the output I'(x,y) depends only on I(x, y).

• [generations5.org] Light on the details, but the chosen images motivate the problem well.


• [cee.hw.ac.uk] A good discussion of the pros and cons of equalization. Poor discussion of discrete HE though.


• [eng.hmc.edu] A good treatment of the discrete case, and a thorough example at the end.

Today, some hints on Assignment 2 were given. Also, binary image processing, including perimeter and area estimation algorithms, were discussed.

Today's lab was mostly Q&A and a chance to provide feedback on the labs.

Sampling theory was covered in today's lab.

Here are a couple of sites that provide further reading on the subject.

• [cs.ucdavis.edu] This site provides a great discussion of sampling theory in terms of what's happening in the frequency domain.


• [egr.uh.edu] This site has a Java applet at the bottom that lets you experiment with different functions and sampling frequencies, and then observe how successfully the original function can be reconstructed.

In today's lab, the first assignment was returned and solutions to the assignment were discussed.

I was asked in lab about posting the solution, but have decided against that. If you'd like a copy of the solutions presented in lab, or have any questions about the assignment in general, send me an email.

The first assignment has been marked; check your grade below.

You can collect your assignment in Tuesday's lab, and I will also briefly cover solutions in the lab. If you have any questions about your mark, see me in lab or send me an email.

Today's lab topic was the Fourier transform (FT). Discussion included the Fourier basis functions, DC component, magnitude vs. phase information, and the convolution theorem. Several image-transform pairs were analyzed.

To familiarize yourself with the concepts of FTs, Assignment 2 is highly recommended.

Assignment 3 deals with image filtering techniques, such as deblurring. The convolution theorem is used in the assignment, so it is useful for learning the practical issues of filtering in the frequency domain.

For further reading on FTs, check these sites:

• [cs.unm.edu] The figures shown in lab today were taken from this site, which provides a nice link between images and their transform.


• [uwa.edu.au] This site discusses some properties of FTs and some simple transform pairs.


• [inf.ed.ac.uk] This page also presents some image-transform pairs. The interesting result is what happens if the phase of a transform is ignored and only the magnitude is used to reconstruct an image.

Covered in lab today:

• Perspective geometry, including projective spaces, homogeneous coordinates, pinhole cameras, affine vs. collinear transformations.


• Homographies: definition, how to compute.


• Tesselations: rectangular vs. hexagonal.


• Pixel neighborhoods: 4N, 6N, and 8N.

The last possible day to submit an assignment is Monday, Dec. 4. Deadlines for all assignments are as follows:

Assignment	Date
Second	Nov. 13
Third	Nov. 20
Fourth	Nov. 27
Fifth	Dec. 4