PCA usage for various ROIs of several images

I have 200 ROIs from each of the 50 images. Hence 10000 ROIs present. For each ROI, I have 96 feature vectors for four different frequency bands(96 X 4 =384). It seems very high dimensional. How to apply PCA for this? How to form data matrix input for PCA? Do I have to apply PCA for each image or each ROI?

[Sumit]

Hey Anitha, r u working in image retrieval ?

Answer 1

[Image Analyst]

I commented on your "Answer" here: my response

What I said there was "It depends on if you want PCA components on each image individually, or the PCA components of the group as a whole."

Answer 2

[Image Analyst]

Indeed 96 feature vectors is a lot for one image. Why do you have so many feature vectors for each ROI? I can see one vector for each ROI, but why does an ROI need 96 vectors? Maybe one vector contains intensity measurements, maybe one contains texture measures, and maybe one contains spatial or shape measures. But 96 of them? What could they all possibly represent?

And how many elements are in each feature vector? Like vector 1 has 5 measurements, vector 2 has 13 measurements, vector 3 has 12 measurements, etc.

PCA will tell you which features are most important, but it seems like you might be able to get some idea in advance of what the important features are and just measure those.

[Anitha Anbazhagan]

Sorry, by mistake I have written 96 feature vectors. Actually it is 96 features from each of the four frequency bands. My features are 32 bin histogram for amplitude, frequency and angle. So 32 * 3 = 96. Four frequency bands are VL, L, M and H. Now help me to apply PCA. Is it me to choose PCA for applying as a whole image or for each ROI? Little confused to reach dimension reduction after processing whole image or after processing each ROI. Please help me out.