We are pleased to announce that the development of the
notorious "Method for Object Recognition of Obscure Nature"
(or Moron, for short) has been switched to the R language.
Given a directory of images (i.e. jpegs), Moron attempts
to predict a category distribution for the content of each
image. The default categories modelled are {healthy,pron,
latex/fetish,japanese_cg,manga,b/w photo}. Supposing
good statistical models can be designed for problems
like these, Moron could eventually be usable as a spam filter
for visual data or to sort images, depending on your taste.
Currently it is not accurate enough for production use.
The Moron package contains a data.frame to estimate a
model to recognize the default categories, and source code
to evaluate new images, collect new training sets (this
can be done simply by putting images of different classes
to different directories and running a few provided functions)
and train new classifiers. The included Local Binary Pattern
(LBP) feature extractors and RGB<->HSV conversion routines
might be of aside interest.
Due to its ability to visualize the predictions made
per image region, Moron is a guaranteed hit in both
parties and classrooms, if only given a juicy enough set
of test images. Even layman (male) audience has been
seen to be captivated by the antics of Moron, efficiently
rivalling traditional aquariums and fireplaces.
Moron is released under GPL. Version 0.6.0b for R can be dl'ed from
http://sourceforge.net/project/showfiles.php?group_id=7434&package_id=68958
The package relies on R libraries 'randomForest' and 'pixmap'.
It has been developed on Linux, but is probably usable on
other systems with only a few simple modifications.
Have fun,
Igor
_______________________________________________
R-packages mailing list
R-packages at stat.math.ethz.ch
https://www.stat.math.ethz.ch/mailman/listinfo/r-packages
