ICDAR’15 Conference. Poster Session 1, Monday, August 24, 2015, 14:40 – 15:40. Foyer 300 , Prouvé Convention Center, Nancy, Lorraine, France.

1. A Multiple-Expert Binarization
Framework for Multispectral Images
Reza Farrahi Moghaddam and Mohamed Cheriet
Synchromedia Lab, École de technologie supérieure (ETS),
University of Quebec (UduQ), Montreal, QC, Canada
ICDAR’15 Conference
Poster Session 1, Monday, August 24, 2015, 14:40 – 15:40
Foyer 300 , Prouvé Convention Center, Nancy, Lorraine, France

2. Summary
• A multiple-expert binarization framework for multispectral images is proposed
• The framework is based on:
– A constrained subspace selection limited to the spectral bands combined with
– State-of-the-art gray-level binarization methods
• The framework uses a binarization wrapper to enhance the performance of the gray-level
binarization
• Nonlinear preprocessing of the individual spectral bands is used to enhance the textual information
• An evolutionary optimizer is considered to obtain the optimal and also some suboptimal 3-band
subspaces from which an ensemble of experts is then formed
• The framework is applied to a ground truth multispectral dataset with promising results
• In addition, a generalization to the cross-validation approach is developed that:
– Not only it evaluates generalizability of the framework
– It also provides a practical instance of the selected experts that could be then applied to unseen inputs

3. The Proposed Framework and Wrapper
Proposed Multiple-Expert
Binarization Framework Binarization wrapper

4. 3-Band Subspace Optimizer
• For each GT image in training set:
– Get the best 3-band associated with the binarization kernel
– Also, add up-to-5 of its ‘tailing’ best 3-bands
• Cumulate all obtained 3-bands of all images
• Rank the 3-bands based on their frequency of
appearance among the training images
• Select most-frequent and most-rare 3-bands to build the
final ensemble of experts:
– Each expert in the ensemble is a 3-band selection combined
with the kernel binarization

5. Results
An Optimizer’s Iteration Example of Performance

6. Results, Cont’d
Performance for the Phase
Congruency (PC) Kernel Sensitivity to the Dataset
a) The Green band (F3) of z92 image of the 10MS dataset. b) Ground truth.
c) Method 1 [DHSA]. d) The proposed framework with the PC binarization
method [13] as its kernel and p = 0.50.

7. Performance in terms of NRM

8. Mohamed
CHERIET, Prof.,
Eng. Ph.D.
Reza
FARRAHI
MOGHADDAM,
Eng., Ph.D.
Principal Investigator Research Associate
mohamed.cheriet@etsmtl.ca imriss@ieee.org,
rfarrahi@synchromedia.ca
Acknowledgement
The authors thank the NSERC of Canada (under Grant RGPDD/451272-13
and Grant RGPIN/138344-14) and the SSHRC of Canada (under Grant 412-
2010-1007) for their financial support.
More Information:
1. PrePrint of this paper on ArXiv: http://arxiv.org/abs/1502.01199
2. ICDAR 2015 MultiSpectral Text Extraction Contest (MS-TEx 2015):
http://www.synchromedia.ca/competition/ICDAR/mstexicdar2015.html
3. Multispectral Datasets: http://www.synchromedia.ca/databases/msi-histodoc
4. Synchromedia Website: http://www.synchromedia.ca
5. Color-To-Gray Converters:
http://www.mathworks.com/matlabcentral/fileexchange/27578
6. Objective Evaluation Tools:
http://www.mathworks.com/matlabcentral/fileexchange/27652
7. DiD – Global Currents (Project): http://txtlab.org/?cat=43
8. Indian Ocean World (IOW) Center: http://indianoceanworldcentre.com/