Moderator: Computer Vision

Tigger
Kernelcompilierer
Beiträge: 404
Registriert: 26. Okt 2007 17:35
Wohnort: Hofheim
Kontaktdaten:

I have a question about the last ToDo in Task2. It says, the image has to be projected on our 20 dimensional subspace. Why is our subspace 20 dimensional? I calculated the subspace from the 760 images in yale_faces/, which resulted in a 760 dimensional subspace.

qgao
Moderator
Beiträge: 33
Registriert: 7. Apr 2009 11:38

### Re: Ex 2 Task 2

Tigger hat geschrieben:I have a question about the last ToDo in Task2. It says, the image has to be projected on our 20 dimensional subspace. Why is our subspace 20 dimensional? I calculated the subspace from the 760 images in yale_faces/, which resulted in a 760 dimensional subspace.
It seems you do not well understand the concepts of space and space basis.

All eigenvectors have the dimension of number of image pixels, and composed the basis of a new space. If you use N eigenvectors as the basis, each image can be represented as a N-D vector. Thus we say the subspace is N dimensional.

Sete
Erstie
Beiträge: 20
Registriert: 18. Mai 2010 14:00

### Re: Ex 2 Task 2

I am a little bit confused by this issue, too. If i am not mistaken, by applying SVD on all the images we obtain 760 eigenvectors, each with a dimension of something about 8000.
So where does the "20" come from? Should we only use the 20 "most important" vectors to form our subspace?

Tigger
Kernelcompilierer
Beiträge: 404
Registriert: 26. Okt 2007 17:35
Wohnort: Hofheim
Kontaktdaten:

### Re: Ex 2 Task 2

Sete hat geschrieben:Should we only use the 20 "most important" vectors to form our subspace?
I think this is it. We choose the 20 dimensional subspace by selecting the corresponding eigenvectors. The idea is clear, i was just a little confused by the formulation of the task.

qgao
Moderator
Beiträge: 33
Registriert: 7. Apr 2009 11:38

### Re: Ex 2 Task 2

Sete hat geschrieben:I am a little bit confused by this issue, too. If i am not mistaken, by applying SVD on all the images we obtain 760 eigenvectors, each with a dimension of something about 8000.
So where does the "20" come from? Should we only use the 20 "most important" vectors to form our subspace?
Right!