Re: ImageJ PCA

Posted by aszia787 on Apr 13, 2007; 2:38pm
URL: http://imagej.273.s1.nabble.com/ImageJ-PCA-tp3699745p3699751.html

Gabriel,

First of all i'm so thankful to you for your patience in answering my questions, i highly appreciate it. Yes, that's true, i get 32 bit images in the stack and i assumed that the first image i get has the maximum eigenvalue. I'll tell you what i am trying to achieve from the PCA. As mentioned earlier, i have a set 56 features/images and i want to extract say the highest 4 eigenvalues. After getting the 4 eigenvalues i will add only the pricipal component planes(4 images) to get the final feature image.

I will try and give you an example. Assume we have 5 images A,B,C,D and E and we pass these images in a stack and extract the PCA. The PCA that we get is in the form of 32 bit images with the highest variance being the first component and so on. My question is, is the first eigenimage that we get, is that the eigneimage of image A or image B? i guess it doesn't tell me that the first image returned is the eigen image of which original feature/image as i have to add the original features with the maximum eigenvalue. I hope i was able to clear the question. If you have any questions, please feel free to ask, i will try to clarify again if needed. Once again thank you so much for your interest.

Regards,

Ali.  

Gabriel Landini wrote

On Friday 13 April 2007 13:01:07 aszia787 wrote:
> I just want to extract the principal
> components. Example, if 14 images were submitted, i wanted to know the
> eigen values of each of the 14 images, and would have selected the first 3
> highest eigen values of the 14 images to have the pricipal component and
> ignore the rest of the features. Is it possible with the BIJ plugin.

Aren't these the greyscale values of the new stack (eigenimages as reported to
the console)?
These are returned as 32 bit. First image is the first component and so on?
As you go along the new stack there should be less explanation the variance
(please correct me if I got this wrong).

Regards,

G.