0

I've got a set of training face images (40 images). Each image size is 28*34. From there, I'm able to get eigenVector, Score, Latent using princomp function in Matlab.

I've got 952 latents (eigenvalues in covariance matrix) which are in descending form : 4.2785 to 0 . Eigenvalues are zeros from k=40 onwards.

May i know what does the the eigenvalues indicate ? (say bigger value means more significant to variance?) how could I identify the best k value (Principal component)?

Thank you so much for your help !

Cape Code
  • 3,584
  • 3
  • 24
  • 45
user3168226
  • 57
  • 1
  • 7

2 Answers2

1

Since you only have 40 input faces you cannot expect to have more than 40 principal components. Therefore the eigenvalue becomes zero for K=40 onwards.
To visualize your results, take the 40 leading eigen vectors, reshape them back to 28-by-34 and imagesc them. What have you got?

Shai
  • 111,146
  • 38
  • 238
  • 371
0
  • EigenValue of an eigenvector represents how important you eigenvector is.
  • Higher the Eigenvalue more important the eigentvector is
  • for 40 images the eigenvectors can not be more than 40
  • You can reconstruct the faces with the eigencvectors to visualize how it looks like
  • The eigenvector with highest eigenvalue is a principal component where most of the data lies
Jai
  • 3,211
  • 2
  • 17
  • 26