SIFT descriptors

[Emeline]

Hello,

I would like to use SIFT histograms to describe my pictures. It works
well but, it takes a really really long time to query images.
And I have read on the following paper:
Features for Image Retrieval: An Experimental Comparison
Thomas Deselaers, Daniel Keysers, and Hermann Ney, 2007

that It takes 0.27 sec to query 10 images in a 10 image database, how
is it possible? Am I missing something on SIFT parameters or FIRE
configuration??

Thanks a lot,

Emeline

[Thomas Deselaers]

Yes!

Very likely you are missing the right configuration. 

Try to use local feature histograms.

Best,

thomas

 

[Emeline]

Thanks for the fast reply!

I think I am already using local feature histograms since I am using
the command : extractsparsesifthistogram...
The options used are : gray pictures and PCA to reduce Sift
descriptors dimension to 32.
So, do you have more clues about what I am missing?

On 3 avr, 13:58, Thomas Deselaers <desela...@gmail.com> wrote:

[Thomas Deselaers]

You probably should try smaller dimensions. 

A histogram with just two bins on each dimension for 32 dimensions already has 2^32 bins and thus is very large (and thus slow to compare). 

Another option to create histograms is to use vector quantization / clustering to get learned histogram bin centers. 

Best,

t.

--
You received this message because you are subscribed to the Google Groups "FIRE - Flexible Image Retrieval Engine" group.
To post to this group, send email to fire...@googlegroups.com.
To unsubscribe from this group, send email to fire-cbir+...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/fire-cbir?hl=en.

--
http://thomas.deselaers.de

[Emeline]

Indeed, it works much more faster with smaller dimension.
Thanks again for the help!