[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [help-GIFT] GIFT Features
From: |
Henning Müller |
Subject: |
Re: [help-GIFT] GIFT Features |
Date: |
Wed, 21 May 2003 10:02:16 +0200 |
User-agent: |
Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.3b) Gecko/20030210 |
Hi,
the four feature groups are:
- color histogram using HSV (18H, 3S, 3V, 4G) and histogram intersection
as distance measure
- Gabor filter response histogram using 3 scales, 4 directions and 10
quantisations, also with histogram intersection
- local color blocks at different scales using an tf/idf-based weighting
(based on frequencies of features)
- local Gabor blocks also using a tf/idf-based weighting
There are various weighting schemes based on tf/idf. You can find some
articles and details on that at
http://viper.unige.ch/research/publiByType.html
When using new distance measures you should write a new weighting
function as a child of the template CWeightingFunction. Wolfgang can
surely give you more details on that. These available weighting
functions can then be transmitted via MRML and chosen in the interface.
If you would like to use a new feature space you need to change the
feature extraction files in the directory FeatureExtraction. There, the
system write the information on how many features there exist and
whether they are from different feature groups which can imply a
different distance measure.
Cheers, Henning
Sailesh Suvarna wrote:
Hi all,
In one of the previous postings ‘Understanding MRML: Purpose of
the"allows-children" ‘
I came across the following…
“
OK. In MRML, you can specify trees of algorithms. This means, you can combine
your algorithm during runtime from blocks. This is done in GIFT/Viper in
order to do the separate normalization thing (four feature groups, that are combined with equal weights or chosen weights as it is done in Monash). “
In the above…what are the “four feature groups” currently used?
-------------
Are the following right for GIFT?
1.At present the HSV colour space is used. (18H, 3S, 3V, 4G)
2.Distance/Similarity measure is calculated using histogram intersection.
--------------
More questions
3.What is the purpose of CQNEuclideanLengthSquare.cc ,if histogram intersection
is the method that is empoyed?
4.If I have to write a plugin say for providing the option of using different
distance measures under a “Distance Measure” Plugin. How should I go about it,
if I would like the user to be able to select different distance measure from a
list of options.
5. Or suppose I’d like to change the colour space used to Luv/Lab and use
Euclidean distance measure, which are the files I could modify/ redo?
Cheers,
Sailesh
_______________________________________________
help-GIFT mailing list
address@hidden
http://mail.gnu.org/mailman/listinfo/help-gift
--
---------------------------------------
Dr. Henning Müller
University Hospitals of Geneva
Division of Medical Informatics
21, rue Micheli-du-Crest
CH-1211 Geneva 4, Switzerland
Tel +41 22 372-6175
Fax +41 22 372-8680
email address@hidden