The number of faces recognized increases with the number of eigenfaces until around 10 or 11 eigenfaces, at which point it flattens out. At this point about 75% of faces are correctly recognized. Fewer eigenfaces require less space, but get worse results in recognizing faces. The best option appears to be 10 or 11 eigenfaces, which gets pretty good results. Using more eigenfaces simply takes more space and didn't improve the results any, and fewer than that number got much worse results.
The face on the right was consistently recognized instead of the one on the left. This seems reasonable as the faces look fairly similar. At 10 eigenfaces, the correct answer appeared second in the results. The difference in MSE was almost 200.
The face on the right was sometimes recognized instead of the one on the left. They do not look much alike besides both showing off their pearly whites so it is not clear why they were such a close match. At 10 eigenfaces, the correct answer appeared second, with difference in MSE of 16. At 11 and 17 eigenfaces, the correct face was recognized.
Me min=.4, max=.6, step=.05
IMG_0031 min=.4, max=.6, step=.05
Group min=.7, max=.9, step=.05
