PROJECT 4

Nathan Day

Recognice Face:

EigenFaces:

Average Face:

Performance of face recognition:

The chart shows that recognition of faces increase steadily from 1 to 11. From 11 onwards, face recognition does not seem to improve with increasing the number of eigenfaces. The computational time still does increase as more eigenfaces improve, but not significantly enough to that it became prohibitive. The best number of eigenfaces to use in this experiment.

Although not all the correct faces were found, the correct face was usually high on the sorted list. Take the first image, for example:

 The correct match is on the left, the actual match was on the right. The seem very similar. 

Find Face:

Here is a cropped image of the elf.tga file:

As you can see, the algorithm actually found the left hand corner of the head in this picture. Unfourtunatly, the correct face match was the second entry in the second face found. I used the parameters specified in the project description.

I also tried to crop my own face:

which resulted in this cropped image with the arguments .44 .55 .01

So, that crop obviously was successful.

Here is the cropped image, IMAGE_31, that were were asked to do. The parameters I used were .52 .52 .02

And lastly, here is a group photo with all four faces found. The parameters used were .97 .97 .02

I found that finding faces seemed to be very finicky. I played around with the parameters until I achieved the results that I was looking for. Additionally, because of the time that it takes to run, it isn't necessarily totally practical to use a large scale range with small increments.