Unsupervised Image Segmentation
ICPR 2014 Contest

Results

Thank you for all submitted segmentations.


There were 15 registered contestants but only six of them submitted their segmentation results. One of them deleted his results before end of the contest. Below is a list of competing methods.

CGCHI
The Combined Graph Cut based segmentation with histogram information on regions method is a combination of global and local coherent information.
Deep Brain Model
Deep brain model is an unsupervised segmentation framework with unknown number of classes simulating the deep structure of the primate visual cortex.
FSEGWWW
A factorizaton-based texture segmenter uses local spectral histograms as features.
VRA-PMCFAWWWAUTHOR1AUTHOR2PPT[1]
The Voting Representativeness - Priority Multi-Class Flooding Algorithm is an unsupervised texture image segmentation framework with unknown number of regions, which involves feature extraction and classification in feature space, followed by flooding and merging in spatial domain. Based on an extended version of [1]. If you use this work, we will appreciate if you cite the paper [1] in your work.
texNCUT
A modification of the NCUT method which is using textural features. Uses prior information about the number of regions.

MW3AR8 - our non-competing method for comparison
An unsupervised multi-spectral, multi-resolution, multiple-segmenter for textured images with unknown number of classes.

During a verification phase there were some problems with execution of codes so the results are slightly varied.

CGCHI
no code/binaries available - we used the submitted results for comparison
Deep Brain Model
conversion from edge images into thematic maps
VRA-PMCFA
using 64-bit version
texNCUT
using newer version (64-bit compatible) of NCUT

After the verification step we computed segmentation results on different mosaic set (using the same preset with different random seeds). It led in three results' groups - submitted, verification and test set. The ranking of the methods stayed unaltered.


And finally, here is the contests' methods ranking.

1. Texture Segmentation Based on Voting of Blocks, Bayesian Flooding and Region Merging
Costas Panagiotakis, Ilias Grinias and Georgios Tziritas (Technological Educational Institute of Crete, Greece)

2. Factorization-Based Texture Segmentation
Jiangye Yuan, Anil M. Cheriyadat (Oak Ridge National Laboratory)

Deep Brain Model
Nan Zhao (Florida State University; Tallahassee, Florida, US)

Combined Graph Cut based segmentation with histogram information on regions
Ruhallah Amandi, Mohammad Farhadi (Zanjan University-Iran)

Textural image segmentation using Normalized Cut
Farzaneh Alizadeh, Nader Karimi, Niloofar Gheissari (Isfahan University of Technology (IUT); Isfahan, Iran)

Below are links to detailed results and their comparisons - tables with numeric values, segmentation images or criteria graphs.