Face Detector with Eigenfaces

View in .ipynb format: link

1. Data

Training Data:

• 3773 images of faces
• excluded images without faces
• A single person may be represented multiple times with varying facial features

Test Data:

• 44 images of faces
• excluded images without faces

2. Result

Selected a random image from the test data and displayed the five best matches in descending order.

Result 1
     

Result 2
     

Result 3
     

Result 4
     

Result 5
     

3. The best (the smallest) dimensional basis vectors (k = 60)

The curve begins to flatten somewhere between 50 and 60. This indicates that beyond 50 or 60 components, additional information captured becomes minimal, suggesting that the image can be effectively reconstructed using approximately 50 to 60 components.

4. Examples of Faces Represented as A Linear Combination of the k Eigenfaces