By Shiyu Luo Dec. 2010

pucefakeAI and Robotics

Nov 30, 2013 (3 years and 6 months ago)

78 views


By
Shiyu

Luo

Dec. 2010


Outline


Motivation and Goal


Methods


Feature extractions


MLP


Classification Results


Analysis and conclusion


References

Motivation and Goal


Oil paintings are of great value


Art


History


Even more counterfeits make it harder to identify the
authentic works


Traditional: signatures, Dates and producers of
canvas, etc.


Proposal: by Digital Image Processing


Brushwork example of one of
da

Vinci’s painting

Left: Brushwork in original painting

Right: micro
-
view of grey
-
degree of the red square

Cont’d


In this pilot project, painting
-
based approaches are
studied


Data set: 8 X
-
rayed paintings from Leonardo
da

Vinci


Method:


Patch selection


Feature extraction


Multi Layer Perceptron

Feature extraction


General requirements:


Intra
-
class variance must be small


Inter
-
class separation should be large


Independent of the size, orientation, and location of the
pattern


Four features are employed


Fourier Transform (Brushworks)


Wavelet Transform (lower resolution image)


Statistical Approach (texture)


E.g.,
2
nd

moment:
a measure of gray
-
level contrast to describe
relative smoothness


Covariance Matrix

Multi Layer Perceptron (MLP)


MLP: Error Back Propagation







A diagram demonstration of Multi Layer Perceptron



Result

Analysis & Conclusion


Generally speaking,
C_rate

can be achieved at
around 40%
-

50%


50x50 patch
-
based generally achieves better and
more stable results than 100x100 patch
-
based does.


For 50x50 patch
-
based, the better and relatively
stable results are those with 6
-
8 neurons in hidden
layer.


Those “excellent” results of 100x100 maybe I’m
“luck” in the 3 trails.

Future work and improvement


X
-
rays maybe one of the limits on achieving better
classification rates; colored paintings could be used
in the future


2
nd

or higher order wavelet transforms maybe used
to improve the feature vector


Other neuron network methods are to be tested to
better suit this painting classification problem

Selected References


Siwei

Lyn, Daniel
Rockmore
, and
Hany

Farid
.
A digital technique for
art authentication
. 17006
-
17010, PNAS, Dec. 2004, vol. 101, no.49.


C. Richard Johnson, Jr., Ella
Hendriks
, Igor J.
Berezhnoy
, Eugene
Brevdo
, Shannon M. Hughes, Ingrid
Daubechies
,
Jia

Li, Eric
Postma
,
and James Z. Wang.
Image Processing for Artist Identification:
Computerized Analysis of Vincent van Gogh’s Painting Brushstrokes
.


Jana
Zujovic
, Scott Friedman, Lisa Gandy,
Identifying painting genre
using neural networks
. Northwestern University.


G. Y. Chen and B.
Kegl
.
Feature Extraction Using Radon, Wavelet
and Fourier Transform
. Systems, Man and Cybernetics, 2007. ISIC.
IEEE International Conference on, pp. 1020
-
1025. Oct. 2007.


Rafael C. Gonzalez, Richard E. Woods.
Digital Image Processing
.
2
nd

edition. Prentice
-
Hall. 2002.