Evan Welbourne
 
Average face and 10 Eigenfaces
 

Correct Predictions vs. Eigenfaces Plot

Questions:
1. The trend in the plot is fairly obvious: An increase in the number of eigenfaces results (roughly) in a corresponding increase in the number of correct recognitions.  This is intuitive because with each added eigenvector, we increase the amount of potential detail achievable via projection onto them.  The trade-off is computational weight vs. accuracy, so to get the best accuracy with the lightest weight computation, we should use about 23 eigenfaces.
 
 
2.  A couple of recognition errors are as follows, recognition could have been better for the first mistake, and was pretty close for the second.  Both errors were made using 33 eigenfaces:

downey.tga smile confused with eckhart.tga, the correct picture of Doug not smiling was ranked 14th.
downey.tga	eckhart.tga	downey.tga

hile.tga smile confused with welbourne.tga, the correct picture of Harlan not smiling was ranked 2nd.
downey.tga	eckhart.tga	downey.tga

back to index