Part I Test of recognition part

The eigen faces we got from the class are shown as following (top 10 eigenfaces):

Based on those eigen faces, the number of recognized faces vs number of eigenfaces is shown as following picture;

The maximum recognition rate is 23/33, about 70%. The recognition rate increases drastically when the number of eigen faces goes up from 1 to 5. Then the increase slows down and reach a steady level at about 70%. In this case, I think 20 eigenfaces might be an optimal choice.

Some faces which can’t be recognized correctly are shown as following:

is always regarded as

is always regarded as

I think the noses of them, shape of eyebrow and lips are same.

Some other examples are:

is always regarded as

For these two, I can’t pick up many similar features between them.

Part II Face detection

For Aseem’s image, I try the parameters on class website, the result is shown in the following picture:

min_scale=0.45, max_scale=0.47, step=0.005;

The face detection of group1 image

min_scale=0.98, max_scale=1, step=0.005

Then I tried the face detection on other images:

For the first picture of Director Granitz, we only get two correct answers of 5,

Three false-positive solutions are hands of the couples and the clothes. I think better add some code concerning the color information. Min_scale is 0.98, Max_scale=1.0, step=0.005.

For the second picture, I got the perfect hit of a person but others are all false positive. That is a grayscale image so that the face similar to the source of eigenfaces might be easy to detect. For the other people, the appearance and expression are far away from the group in the class. Min_scale=0.52, Max_scale=0.54, step=0.005.