Eigen Faces

Eigen Faces

Average Face

Eigen Face 0	Eigen Face 1	Eigen Face 2	Eigen Face 3	Eigen Face 4
Eigen Face 5	Eigen Face 6	Eigen Face 7	Eigen Face 8	Eigen Face 9

Using 10 eigen faces my recognition rate was 80% (20/25).

From the plot we can see that recognition rate increases as
the number of eigen faces used increases, until around 10 eigen
faces are used. After that there is no increase in recognition rate.
This suggests that 10 eigen faces is a good number to use for this
data set.

A couple of the images were not recognized consistently:

		01.tga
		07.tga
		11.tga
		14.tga

Perhaps these images were not cropped as consistently as the
others. Notice the difference of the amount of forehead showing
between the smiling and non-smiling pictures above as compared to
those below, which were fairly consistently recognized.

		01.tga
		03.tga

My program seems to do all right on finding single faces and multiple faces
that are all oriented toward the camera. But it seems to fail on faces of
different sizes and ones that are oriented away from the camera.

For a user database using 6 eigenfaces with this dataset I decided
the best threshold for verifyface is:
0.002425
I decided on this threshold using recognizeface (uses the same underlying
logic), because it can output if the match is correct and what the mse
(threshold) is for an except and for a reject (by looking at the mse for the
highest and second highest matches). I ignored the 7 cases where it recognized
the wrong face, but I included those cases in the statistics below:

case	instances	rate
total positives	17
total negatives	7
false positives	2	29%
false negatives	2	29%

The only extra credit I did was to implement the speed up to calculating
the eigen vectors, as described on the project page.