Disciplines
Computer Sciences (90%); Arts (10%)
Keywords
PAINTING ANALYSIS,
NEURAL NETWORKS,
IMAGE PROCESSING,
IMAGE CLASSIFICATION
Final report
The goal: The main aim of this project was to find the painter of an image, using only a computer.
The start: We started with more than 500 drawings and paintings from twelve artists (three children included).
These images were scanned to get digital copies.
The method: We used a step by step procedure:
- analysis of general and structural properties of an image
- detailed analysis of individual objects, especially of lines in an image
- finding best features for classification
- other important properties of an image
All together we calculated 84 features for every image
General results:
Several theoretical considerations and practical experiments suggested using finally only 4 - 6 features for
classification, namely:
Color properties (hue), fine/coarse (fractal dimension), structure (edge density),
organisation (cooccurrence).
Our method gave a success rate of at least 85 % to find the painter of an image.
Particular results:
Extraction of lines from an image (this is difficult if the lines have crossings; how to continue a line after a
crossing ?). We developed a solution which satisfied almost all of our demands. We finally incorporated also lines
into the classification process; this gave better results for images with typical lines (as lines are an important aspect
of the personality of an artist).
What is the most interesting part of an image? We developed a method to find the most interesting region in an
image using variance.
Similarity of images: Using the features mentioned above we get a natural measure for the similarity of images.
Conferences and Publications:
Presentation at three international conferences, four publications
Public presentation at unihautnah, Nov 4 - 6, 2004 (the university of Salzburg presented six research projects),
details: http://www.unihautnah.sbg.ac.at/mathematik/index.html
Future:
Many results of our project research should be interesting for schools and for musea of arts: Perhaps this will help
more young persons to get interested in arts (if they can play with a computer and images).