The experimental results of the project are separated in 2 parts:
 

Key point Localization:

To test Key point localization, 10 pictures of several people were taken from the database. We tried to find 3 key points:

    - The inner side of the left eye

    - the tip on the nose

    - the right end of the lip

A Matlab implementation of our key point detection algorithm was applied, and the resulting key points were highlighted such that key point results can be validated by visual inspection.

The results for the 3 features are the following:

neye - 80%

nnose - 80%

nlip   - 20%

n

 

 

 

 

 

 

 

Classification:

To test classification, the underlying database was used. The coordinates of all 20 Key point associated with 1020 images of 11 persons were extracted, and the resulting 40-dimensional vectors were divided into a testset and a trainingset. Our Nearest Neighbor classification approach yielded satisfying recognition rates on the test set: for

98 %

of all images, the person was identified correctly.