EE 576

Project 3

Reed Lane

 

Extra Credit:  todo 5 verifyFace  was implemented successfully.

 

Note: the TODO’s in eigenfaces.cpp and faces.cpp were filled in and eigfaces.h was slightly modified.

 

 

Part 1

 

 

 

Question 1)

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

 

The plot show that using more eigenfaces give a better result.  But there is a limit to how many you need.  After 11 eigenfaces the results don’t get any better with more eigenfaces.   You should use around 10 or 11 to recognize a face.  Any more is a waste of omputation.

 

Question 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?

 

Errors did occur in this experiment.  Certain faces were never recognized with any amounts of eigenfaces. 

For example Ko was recognized as Zhang, and the correct answer didn’t even appear in the top 5.   But Zhang was recognized correctly as himself.

Ko             Zhang

 

 

 

 

part 2

 

When cropping Aseem’s face the result was good.  Their was no mistake.  I used the scale numbers given in the project description.

Min scale:  0.45    max scale:  0.55     step of scale:   0.05

 

When I tried to crop my face from this picture I got a false positive in the whiteboard.  This is one of those low variance areas.  Even multiplying by the difference from the average face and dividing by the varience couldn’t get rid of the false positive.

Min scale:  0.45    max scale:  0.55     step of scale:   0.05



For the group of three students from class the result is good.  Here the formula factoring in the variance and distance were essential to getting only the faces with no false positives or negatives. 

Min scale:  0.8    max scale:  0.9     step of scale:   0.1

 

 

This picture has very poor results.  I found that when the faces in the picture are smaller than the faces used for the database the program was not good at finding the faces.  I think the variance factors too much into the error in this case.

Min scale:  1.2      max scale:  1.4     step of scale:   0.1