spacer.png, 0 kB
  spacer.png, 0 kB
spacer.png, 0 kB
spacer.png, 0 kB
HEp-2 Cells Classifier PDF Print


  • Roman Stoklasa
  • Tomáš Majtner
  • David Svoboda
  • Michal Batko


This program is an implementation of a k-nearest neighbor classifier build on top of the MESSIF framework and i3d image processing library.
Detailed information about this software can be found in the official method description document.

We developed two versions of the classifier.

Version v1.1.1 is implemented for Windows OS, but can be compiled also on Linux OS. It uses following global image descriptors:

  • Local Binary Patterns (LBP)
  • Haralick features
  • Color Structure (from the MPEG-7 descriptors)
  • Granulometry-based descriptor
  • Surface description

Almost all descriptors (except Color Structure) were implemented with help of i3d image processing library.

Version v1.2.1 of the classifier is an enhancement of the version v1.1.1, which utilizes also local descriptors besides the global ones. This version is implemented for Linux OS only.

The classification process of each image can be divided into four stages:

  • preprocessing - the aim of this stage is to normalize images and enhance them before image descriptors are used.
  • k-NN search - neighbors in the k-NN query is searched according to some aggregated distance function, which combines several different descriptors with different weights.
  • joining information from all nearest neighbors and inferring of classification estimate
  • combination of partial classification estimates and computing final classification

HEp-2 Cells Classification Contest

The software was developed for the Contest on HEp-2 Cells Classification, which was hosted by the 21st International Conference on Pattern Recognition (ICPR) in 2012.

The task was to design and implement a pattern recognition system able to classify the pre-segmented cells belonging to HEp-2 images in one of the following pattern classes:

HEp-2 Cells
Examples of HEp-2 Cells images (after some preprocessing for better visualisation)


Both classifiers were tested on the test dataset and have the accuracy of about 95%.
Detailed information about this software can be also found in the official method description document.

If you are interested in these classifiers and you want to know more about it or to try it by yourself, do not hesitate to contact the main authors:

  • Roman Stoklasa's email: Email address of Roman Stoklasa
  • Tomáš Majtner's email: Email address of Tomáš Majtner


This software is the result of the project P302/12/G157 Dynamics and organization of chromosomes during the cell cycle and the differentiation of norms and pathology provided by the Grant Agency of the Czech Republic and the project MUNI/A/0914/2009 Student Project Grant at MU (specific research, rector's programme).

This result is consistent with the objectives of the project. The owner of the result is Centre for Biomedical Image Analysis (CBIA) at Masaryk University, a public university, ID: 00216224. After contacting the authors and describing the motivation of usage the software, CBIA will allow to use some components or entire software free of charge and without territorial restrictions for academic & non-commercial use in usual way, that does not depreciate its value. This permission is granted for the duration of property rights.

This software is not subject to special information treatment according to Act No. 412/2005 Coll., as amended. In case that a person who will use the software under this licence offer violates the licence terms, the permission to use the software terminates.

Written by Roman Stoklasa   
Last Updated ( Monday, 22 April 2013 )
spacer.png, 0 kB
spacer.png, 0 kB
spacer.png, 0 kB