 Segmentation and supervised classification of image objects in Epo dopingcontrol
 Determining the confidence interval in fitted measured data by the regression line
Result 1: Segmentation and supervised classification of image objects in Epo dopingcontrol
Projects: 2/7087/27 a 1/0077/09
Authors: I. Bajla, F. Rublík, B. Arendacká, I. Farkaš, K. Hornišová, S. Štolc, V. Witkovský
Some years ago the World Antidoping Agency (WADA) introduced new criteria of test positivity in Epo doping control. In the framework of an international WADA project GASepo, these criteria lead to the requirement of extending the system of image analysis by mathematical methods of segmentation and classification. The proposed classification method is a significant contribution to the classification topic. It is based on weighted ranks (WR) of the values of a criterial function that involves Mahalanobis distances generated by the given data populations. We have modified the initial decision rule for two classes and proposed a generalized model of multiclass clasification. This result exceeds the frame of the Epo doping and it is significant also for other application areas in which the normal distribution assumptions are critically violated. The WR classifier proposed in our paper has been evaluated on real samples of segmented objects of Epo images. They have been compared to three selected well known classifiers: Fisher linear classifier, Support Vector Machine, and Multilayer Perceptron. The proposed WR classifier manifested the best results in majority of the cases having been explored.
Cooperation with universities: Faculty of mathematics, physics and informatics, Comenius University, (Assoc.Prof.Dr.I.Farkaš)
Foreign partner: Austrian Research Centers GmbHARC Seibersdorf,
The presented result made possible development of optimum versions of particular algorithms for classification of objects in Epo images, implemented in the GASepo system. This system is used in everyday practice of most of the WADA accredited doping control labs worldwide.
The illustration of an Epo image with segmented objects and artefacts, which have to be separated using classification.
Publications:
 BAJLA, I.  RUBLÍK, F.  ARENDACKÁ, B.  FARKAŠ, I.  HORNIŠOVÁ, K.  ŠTOLC, S.  WITKOVSKÝ, V.: Segmentation and supervised classification of image objects in Epo dopingcontrol. Machine Vision and Applications 20 (4), 2009, 243259, DOI 10.1007/s0013800701200.
 ŠTOLC, S.  BAJLA, I.: Improved accuracy of band detection in GASepo system for quantitative analysis of images in Epo doping control. Measurement Science Review 7 (1), 2007, 1418.
 ŠTOLC, S.: Segmentation and Classification of Objects in Images Used in Epo Doping Control. PhD thesis. 39529 Bionics and Biomechanics, Faculty of Mechanical Engineering, Technical University, Košice, 2009.
 ŠTOLC, S.  PENZ, H.  MAYER, K.  HEISSCZEDIK, D.: Verfahren zur ermittlung von Helligkeitswerten. Austrian patent AT 501 763 B1. Austrian Research Centers GmbH  ARC Seibersdorf, 2006.
Result 2: Determining the confidence interval in fitted measured data by the regression line
Projects: APVV RPEU000806, VEGA 2/7082/27, VEGA 1/0077/09, VEGA 2/0019/10, VEGA 2/0201/10, MŠMT ČR LC 06024
Authors: G. Wimmer, K. Karovič, Witkovský, V.
At the two dimensional microstructures metrology by the length photoelectric comparator are the structure edge coordinates evaluated from this part of photoelectric signal, which may be fitted by a straight line. It is used to evaluate the edge coordinate at the given photoelectric signal threshold value by the estimation of straight line regression coefficients which approximate the linear part of the photoelectric signal. The standard deviations of edge coordinates, structure position and width are evaluated by the "classic" low of the uncertainty propagation by the calculation of physical quantities.
Our proposed method for computing the structure edge coordinate involves the confidence interval round the regression line in which are expected with the given probability P=1α; (α≤1) the photoelectric data. We have evaluated the interval borders, in which are expected with the given probability the true edge coordinate, structure position or width. This procedure is suitable in the Metrologic institutes able to microstructure dimensional measurements, needed e.g. for the microelectronic.
Figure: Fotoelektric signal from reflex line and confidence intervals in which (with 95% probability) lie real values of coordinates of the line rims and the line axis in a PTB nanocomparator. For illustration, the width of these intervals for the reflex line with rectangular fotometric profile of a nominal width of 4 μm is of the order of nanometers.
Publications:
 KÖNING, R., KAROVIČ, K., WIMMER, G., WITKOVSKÝ, V.: Estimating the standard uncertainty contribution of the straightline fit algorithm used to determine the position and the width of a graduation line. Metrologia 49 (2012) 169179,
 WIMMER, G.  KAROVIČ, K.: Determining the confidence interval for the center and width of a structure in fitting measured data by the regression line. In: MEASUREMENT 2009. 7th International Conference on Measurement. Bratislava, Institute of Measurement Science SAS, 2009, 4548,
 WIMMER, G.  KAROVIČ, K.: Interval estimators of the centre and width of a twodimensional microstructure. Measurement Science Review, 9, 2009, 4, 9092.
