CS/EE 576: Project 3 -- Eigenfaces

By Zhi Zhou, 05/18/2003

Experiments Results

Testing recognition with cropped class images

Procedure

1. Use the cropped, non-smiling students (in class_nonsmiling_cropped) to compute 10 eigenfaces.

The average face:

10 eigenfaces:

2. Use the same set of images to compute a userbase.

3. Have the program recognize the cropped, smiling students.

A.Experiment with the number of eigenfaces used. Try using the mean face plus 1 through 33 eigenfaces, at a granularity of 2.

B.Plot the number of faces correctly recognized versus the number of eigenfaces used:

Questions

1.Describe the trends you see in your plots. Discuss the tradeoffs; how many eigenfaces should one use? Is there a clear answer?

We can see, generally the more eigenfaces are used, the more faces can be correctly recognized. Using a large data set should improve the recognization rate, with the claculatoin expense.

To get the best result with least eigenfaces, 21 eigenfaces should be taken. However, there is not a clear answer since the result could vary with different eigenfaces.

2.You likely saw some recognition errors in step 3; show images of a couple. How reasonable were the mistakes? Did the correct answer at least appear highly in the sorted results?

Wrong recognition 1:  

The 2nd match is correct.

Wrong recognition 2:  

None of the first 5 matches is correct, which means the correct face does not necessarily appear highly in the sorted results and MSE of coefficients is not so effective.

Cropping and finding faces

Procedure

1.Use the 10 eigenfaces file computed in the previous problem.

2.Use your program to crop the aseem.tga image. Use min_scale,max_scale, step parameters of .45, .55, .01.

By min_scale,max_scale, step parameters of 0.45, 0.55, 0.01.



3.Find a digital picture of yourself.

My friend Jian: by min_scale,max_scale, step parameters of 0.32, 0.36, 0.1.



4.Experiment with min_scale, max_scale, and step parameters to find ones that work robustly and accurately.

5.Find the faces in two different photos (use the crop=false option)

A.First, try the group1.tga image given.

I took the parameter of 0.9, 1.1, 0.1:



B.Find and try another group photo. This could be of family and/or friends, or one from the web. The photo should have at least four faces.

I took the parameter of 0.9, 1.0, 0.1:



(Source: http://www.vpsa.vt.edu/images/Vpsa%20Photos/Vpsa%20Group.JPG)

Questions

1.What min_scale, max_scale, and scale step did you use for each image?

See above.

2.Did your attempt to find faces result in any false positives and/or false negatives? Discuss each mistake, and why you think they might have occurred.

In the picture, the gentleman's face is not frontal, which makes the recognition difficult. And the other two false positions should be due to the low-texture areas.

Extra Credit:

1. Speedup in calculating eigenvectors

2. Verify Face

Procedure

1.Use the cropped, non-smiling students to compute 6 eigenfaces.

2.Use the same set of images to compute a userbase.

3.Have the program verify the cropped, smiling student images in the smiling userbase. Test by verifying each student against his or her smiling face, as well as each student against a smiling face which is of someone else.

4.Experiment with various thresholds to find the one that gives the fewest false positives and fewest false negatives

Questions

1.What MSE thresholds did you try? Which one worked best? What search method did you use to find it?

For the MSE thresholds, I tried 60000, 30000, and 20000, in which 20000 worked best. I monitored those MSEs and set threshold correspondingly.
When threshold=60000, the false negative rate = 18.75%, and the false positive rate=0.

2.Using the best MSE threshold, what was the false negative rate? What was the false positive rate?

When threshold=20000, the false negative rate = 3.13%, and the false positive rate = 0.