Re: [help-GIFT] Searching for similarities

[MUELLER Henning]

Hi,

a very good database to try out the search for duplicates is the
TSR10000 database which contains most images two or three times, but
with some slight differences, especially with different colours as they
were on an analog tape and were grabbed with a framegrabber.
If you do a query with all features you can see that the "same" images
get similarity scores between 0.8 and 0.9 which does get worse when
using colour blocks and histogram, but much much better, when you use
only colours histogram.
You can also see that the difference between a duplicate and all the
other images is quite large.
By the way, the normalization ist not really correct when you do not use
all the feature groups. The image itself then gets a score of 0.5 or
0.25.

I think a colour histogram with an even smaller number of colours would
even work better, depending on the kind of images you have.
If you not only looking for almost-the-same-duplicates but also images
that are part of another image in the database, matching local color
histograms might work better.

Cheers, Henning
-- 
     ----------------------------------------------------------
     Henning Mueller, Computer Vision Group
     Computer Science Department, University of Geneva
     24, rue du General Dufour, CH-1211 Geneva 4, SWITZERLAND  
     Phone : +41(22)705 7633; fax: +41(22)705 7780
     address@hidden
     ----------------------------------------------------------