Institute of Information Theory and Automation

Mgr. Jiří Boldyš, Ph.D.

Jiří Boldyš
Position: research fellow
Department: Image Processing (ZOI)
Research interests: pattern recognition, surveillance cameras, image and video restoration, particularly inpainting, moment invariants, invariants to convolution, physics and image processing in nuclear medicine, wavelet transform, thin film image processing
Publications at UTIA: list

Email:
Phone: +420 26605 2247
Fax: +420 284680730
Room: 34
Address: Pod Vodárenskou věží 4, CZ-182 08, Praha 8, Czech Republic

Details:

WORK EXPERIENCE
1998 – now
Research fellow (mostly part-time)
Institute of Information Theory and Automation, Academy of Sciences of the Czech Republic, Prague, Czech Republic

2009 – now
Self-employed
- simulations and modelling

2009
Production planning specialist
ŠKODA AUTO, a.s.
- innovations

2006 – 2008
Developer in the R&D team
UPEK, s.r.o.
- development and implementation of biometrics algorithms

2004 – 2006
Researcher (post-doc)
Laboratory Informatique, Image, Interaction at the University of La Rochelle
- algorithms for corrections of film defects

1999 – 2004
Researcher
Media Technology Laboratory at the Helsinki University of Technology

1998
Physics teacher
Institute of Chemical Technology in Prague, Czech Republic

EDUCATION AND TRAINING
1996 – 2004
PhD
Faculty of Mathematics and Physics, Charles University in Prague
- Mathematical and computer modelling
- Thesis: Multiscale Semicontinuous Thin Film Image Analysis

1991 – 1996
Master’s degree
Faculty of Mathematics and Physics, Charles University in Prague
- Physical electronics and vacuum physics
- Thesis: Diagnostics in chemically active plasma

AWARDS
2003: Best publication of the Institute of Information Theory and Automation, Academy of Sciences of the CR, in 2003
Successful leading of diploma students - two first prizes from two different students, see below



Research projects:


Monte Carlo simulation of PET images for injection dose optimization
When a patient is examined by positron emission tomography (PET), radiotracer dose amount (activity) has to be determined. However, the rules for activity correction according to patients’ weight used nowadays do not correspond with practical experience. Very high image quality is achieved for slim patients, whereas noisy images are produced for obese patients. There is opportunity to modify the correction rule with the aim to equalize image quality within the broad spectrum of patients and to diminish radiation risk to slim patients, with special importance for children.
We have built a model of a particular PET scanner and approximated human trunk, which is our region of interest, by a cylindrical model with segments of liver, outer adipose tissue, and the rest. We have performed Monte Carlo simulations of PET imaging using the GATE simulation package. Under reasonably simplifying assumptions and for special parameters, we have developed curves that recommend amount of injected activity based on body parameters to give PET images of constant quality, the quality being expressed in terms of noise equivalent counts. The dependence qualitatively differs from the rules used in clinical practice nowadays, and the results indicate potential for improvement.

Invariants to symmetrical convolution
We can derive invariants to convolution with a symmetrical kernel in an arbitrary dimension. They are expressed in the Fourier domain as a ratio of the Fourier transform and of the symmetrical projection of the Fourier transform. In 2D and for dihedral symmetries particularly, we have expressed the invariants as moment forms suitable for practical calculations.

Substitution rules to easily construct moment invariants, e.g. to blur
We have shown that simple substitution rules can be used to obtain combined moments invariants, e.g. to both geometric degradation and blur.

Moment invariants to rotation and blur in arbitrary number of dimensions
We have presented the construction of combined blur and rotation moment invariants in arbitrary number of dimensions. Moment invariants to convolution with an arbitrary centrosymmetric filter have been derived, and then their rotationally invariant forms have been found by means of group representation theory to achieve the desired combined invariance.

Exemplar-based inpainting
Our field of concern is movie restoration, particularly scratch concealment. The focus was on a single frame (still image) inpainting. We used exemplar-based approach, which uses patches from the known areas and copies their content to the damaged area.

Thin film description by wavelet coefficients statistics
Descriptive and robust features based on wavelet transform coefficients have been proposed for a multiscale thin film image analysis. The features are based on one- and two-dimensional histograms of the wavelet transform coefficients and they can be calculated for every scale of the wavelet decomposition. A one-dimensional histogram model is extended to describe also two-dimensional histograms, by means of calculating marginal histograms and by sampling the two-dimensional histograms in orientation.



Supervised students:


Jiří Novotný, 2014
Analysis of results of particle phenomena modeling in computer games
Department of Software and Computer Science Education, Faculty of Mathematics and Physics, Charles University in Prague

Petr Bulušek, 2013
Visually realistic modeling of dynamic objects deformations
Mathematical Institute of Charles University
Abstract:
The present work deals with simulation methods for rigid bodies and deformable bodies. In the rst chapter you can nd research of some methods for simulation of rigid body physics with emphasis on method used in open source physics engine Bullet. In second chapter you can nd methods for simulation of deformable bodies, again with emphasis on Bullet physics engine. In last chapter model order reduction technique is presented. This method enables to reduce system of ordinary di erential equations. These equations come for example from applying nite element method to partial di erential equations describing motion of elastic body. The technique is studied on bar truss systems.

Jan Vyhnánek, 2012
Signal complexity evaluation in the processing of functional magnetic resonance imaging
Department of Software and Computer Science Education, Faculty of Mathematics and Physics, Charles University in Prague
Abstract:
Functional magnetic resonance imaging has been recently the most common tool for examining the neural activity in human and animals. The goal of a typical data-mining challenge is the localisation of brain areas activated during a cognitive task which is usually performed using a linear model or correlation methods. For this purpose several authors have proposed the use of methods evaluating signal complexity which could possibly overcome some of the shortcomings of the standards methods due to their independence on a priori knowledge of data characteristics. This work explains possibilities of using such methods including aspects of their con guration and it proposes an evaluation of performance of the methods applied on simulated data following expected biological characteristics.The results of the evaluation of performance showed little advantage of these methods over the standard ones in cases when the standard methods were possible to apply. However, some of the methods evaluating signal complexity were found useful for determining the regularity of signals which is a feature that cannot be assessed by the standard methods. Optimal parameters of the methods evaluating signal regularity were determined on simulated data and nally the methods were applied on the data examining emotional processing of subjects with bipolar disease. It has been shown that examining the signal regularity can point at areas with possibly worsened emotional processing in bipolar pacients.

Michal Srna, 2011
Registration of NM and CT images with low-dose CT-based refinement
Faculty of Biomedical Engineering, Czech Technical University in Prague
Abstract:
Functional SPECT image indicates activity distribution of radiopharmaceutical, which is being accumulated i.e. in malign tumors and their metastasis. In anatomical CT image they may not be easy distinguishable from healthy tissue.
We have access to the multimodal imaging system SPECT/CT, which provides functional SPECT image simultaneously with anatomical low-dose CT image with low information density. Both of images have identical geometry and are already registered. Diagnostic CT imaging system is also available providing anatomical image with high information density.
This work deals with description, implementation and successful testing of elastic registration method thin-plate splines for low-dose CT and diagnostic CT image registration. Low-dose CT image is used only for transformation calculation, which is consequently applied on geometrically identical SPECT image.
Image fusion of registered SPECT and diagnostic CT acquired by implemented SW provides better localization possibilities of pathological processes indicated by SPECT image.

Jiří Dvořák, 2010
"Best diploma thesis in Mathematics in 2009/2010" awarded by the Dean of the Faculty of Mathematics and Physics, Charles University in Prague
Influence of injection dose and body parameters on PET image quality by means of Monte Carlo simulations
Department of Probability and Mathematical Statistics, Faculty of Mathematics and Physics, Charles University in Prague
Abstract:
Positron emission tomography (PET) is an imaging technique allowing to determine radiotracer distribution in a patient’s body. This work reviews basic principles of PET imaging. It also uses the random field theory to detect locations with increased radiotracer uptake. This procedure is tested on a collection of simulated PET images.
The aim of this work is to describe the quality of simulated PET images in terms of both the patient’s physical parameters and the amount of applied radiopharmaceutical. The relations are used to provide curves of constant quality determining the amount of radiopharmaceutical needed to achieve desired quality of the resulting images. The resulting curves are compared with the formula currently used in medical practice.

Jan Kratochvíla, 2009
First prize in the competition of ABRA Software, category Business Intelligence
Time Analysis of 3D Data in Nuclear Medicine
Department of Physical Electronics, Faculty of Nuclear Sciences and Physical Engineering, Czech Technical University in Prague
Abstract:
This Master’s Thesis deals with comparison of two functional volumes in nuclear medicine. Their differences are visualized fused with an anatomical image in
order to find epileptic seizures foci in the brain.
The theoretical part contains a description of modalities used (PET, SPECT, and MRI), epilepsy, the ITK and FLTK toolkits, and the method for image registration.
An analysis of four published methods dealing with the topic was carried out. They are SISCOM, B.R.A.S.I.L., BRASS, and Zubal’s work. All methods are based on the same framework – input data are registered and then the functional studies are normalized and subtracted voxel by voxel. Differences are finally visualized fused with the anatomical study.
Based on this analysis, a new program called Diagnomed was implemented. Its main goal was simplicity of use for clinical workers. Diagnomed was tested in cooperation with a physician on four real patients and provided expected results.

Current Ph.D students:

List of projects:

Responsible for information: ZOI
Last modification: 11.11.2015
Institute of Information Theory and Automation