Fusion of BOW descriptor and Zernike Moments Descriptor

asked 2017-11-14 02:35:31 -0500

Dipak gravatar image

updated 2018-04-21 06:51:04 -0500

I am trying to write a CBIR system, i tried BOW with SIFT descriptor method but it is not efficient enough, Then i tried Zernike Moments descriptor. Individually they are not up to the mark for my project. So i am thinking to use both of them and create some kind of fused descriptor which will i am sure give better result.

Can some one point me to right direction on how to use global and bow descriptor together for CBIR system.

Description of the Project We have millions of news paper ads and we need to query the database and find out similar ads your help will be deeply appreciated

edit retag flag offensive close merge delete



Hi, after a long time i found an article looks promising but i could not grasp the article in entirety. I am posting this here if anyone can help to decode it and explain it in simple language thank you in advance link text

Dipak gravatar imageDipak ( 2018-04-21 06:50:55 -0500 )edit

how long are your zernike descriptors, and how do you acquire those ?

sure, fusing descriptors makes sense quite often, given you carefully weight them. maybe a bow approach for the zernike ones, too ? (they were quite long, iirc)

berak gravatar imageberak ( 2018-04-21 07:44:29 -0500 )edit

ZM is 37 Float and SIFT is more than 1000 how these can be fused and BOW can be created...

Dipak gravatar imageDipak ( 2018-04-23 08:09:02 -0500 )edit

then, i guess, it does not make much sense to fuse them, because the will only get outnumbered by your SIFT BOW vecs.

berak gravatar imageberak ( 2018-04-23 08:18:46 -0500 )edit