DT-Binarize - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 146

CHAPTER 12

DT-Binarize

A decision tree based binarization for protein

crystal images

İmren Dinç 1 ; Semih Dinç 1 ; Madhav Sigdel 1 ; Madhu S. Sigdel 1 ; Ramazan S. Aygün 1 ; Marc L. Pusey 2

1 DataMedia Re-

search Lab, Computer Science Department, University of Alabama Huntsville, Huntsville, AL, USA
2 iXpressGenes Inc., Huntsville, AL, USA

Abstract
In image thresholding problems, there are some cases that single thresholding technique may not gen-
erate good binary images for all samples. Using multiple methods may help to overcome this limitation,
but this idea brings another problem. It is not a trivial task to select proper thresholding method for
each image in the dataset. In this study, we propose a generic framework for image thresholding that
utilizes a tree based structure to determine the best thresholding approach for a particular case. We call
our method “DT-Binarize,” and apply our method to the protein crystallization image dataset. In our
experiments, we compare the results with the reference binary images that are manually generated by
our research group. In order to provide more reliable and objective comparison, numerical results are
presented as well as the visual results. Experimental results indicate that the correctness of DT-Binarize
outperforms other methods by 10% on the average.

Keywords

Global thresholding
Image binarization
Protein crystallization
Decision tree

Acknowledgment
This research was supported by National Institutes of Health (GM090453) grant.