Does It Make Sense to Develop New Feature Selection Methods?
Somol, Petr; Novovičová, Jana
2007 - English One of hot topics discussed recently in relation to pattern recognition techniques is the question of actual performance of modern feature selection methods. Feature selection has been a highly active area of research in recent years due to its potential to improve both the performance and economy of automatic decision systems in various applicational fields, with medical diagnosis being among the most prominent. Feature selection may also improve the performance of classifiers learned from limited data, or contribute to model interpretability. The number of available methods and methodologies has grown rapidly while promising important improvements. Yet recently many authors put this development in question, claiming that simpler older tools show to be actually better than complex modern ones -- which, despite promises, are claimed to actually fail in real-world applications. Jedno z aktuálních témat diskutovaných v současné době ve vztahu k oboru rozpoznávání je otázka skutečné účinnosti moderních metod výběru příznaků. Výběr příznaků je stále zkoumaná oblast neboť může zlepšit jak účinnost tak i hospodárnost automatických rozhodovacích systémů v mnoha aplikačních oblastech, z nichž mezi nejdůležitější patří lékařská diagnostika. Výběr příznaků může také zlepšit účinnost klasifikátorů, navržených na základě omezeného množství dat, nebo přispět k interpretaci modelů. Zejména poslední dobou bylo vyvinuto mnoho metod a metodologií slibujících významné zlepšení. Nicméně objevila se také řada kritických příspěvků prohlašujících, že jednoduché staré nástroje jsou ve skutečnosti lepší než složité moderní metody, které, navzdory slibům, selhávají v reálných aplikacích. Ve zprávě zkoumáme toto tvrzení, ukazujeme několik ilustrativních příkladů, vyvozujeme závěry a doporučení týkající se očekávané účinnosti metod výběru příznaků. Keywords: feature selection; subset search; search methods; performance estimation; classification accuracy Available at various institutes of the ASCR Does It Make Sense to Develop New Feature Selection Methods?

One of hot topics discussed recently in relation to pattern recognition techniques is the question of actual performance of modern feature selection methods. Feature selection has been a highly active ...

Somol, Petr; Novovičová, Jana
Ústav teorie informace a automatizace, 2007

Limit Laws for f-disparity Statistics under Local Alternatives
Vajda, Igor
2007 - English Limit Laws for f-disparity Statistics under Local Alternatives Keywords: disparity statistics; asymptotic distribution Available at various institutes of the ASCR Limit Laws for f-disparity Statistics under Local Alternatives

Limit Laws for f-disparity Statistics under Local Alternatives

Vajda, Igor
Ústav teorie informace a automatizace, 2007

Hidden Periodicity in Interpolated Signals and in Their Derivatives
Saic, Stanislav; Mahdian, Babak
2007 - English Many applications like, texture analysis, image registration, etc., are based on interpolation. Therefore, without the detailed knowledge of how the statistics of the signal is changed by the interpolation process, applications based on statistical approaches working with interpolated signals or with their derivatives can yield miscalculations and unexpected results. Therefore a detailed knowledge about the changes brought into the signal by interpolation could be critical. Sometimes, there exists a need to examine whether a signal has been interpolated. Řada aplikací jako texturní analýza, registrace obrazů a další jsou založeny na interpolaci. Detailní informace o změně statistických vlastností interpolovaného signálu a jeho derivací může pomoci při identifikaci zda a jak byl zkoumaný signál interpolován. Keywords: image processing; Periodicity; image forgery Available at various institutes of the ASCR Hidden Periodicity in Interpolated Signals and in Their Derivatives

Many applications like, texture analysis, image registration, etc., are based on interpolation. Therefore, without the detailed knowledge of how the statistics of the signal is changed by the ...

Saic, Stanislav; Mahdian, Babak
Ústav teorie informace a automatizace, 2007

Periodic Properties Of Resampled Images
Mahdian, Babak; Saic, Stanislav
2007 - English In recent years, due to the advent of high-performance commodity hardware and improved human computer interfaces, it has become relatively easy to create fake images. Modern, image processing software enables forgeries that are undetectable by the naked eye. Typically, to create high quality and consistent forgeries, several types of tampering techniques are employed simultaneously. One of the most often used techniques is resampling. When two or more images are spliced together, in order to create a consistent tampering, almost always geometric transformations such as scaling or rotation are needed. In this work we propose an effective method to automatically detect resampled portion of images. Generally, resampling is based on an interpolation procedure. We study specific periodic properties presence in interpolated signals and use this for our detection method. We demonstrate our results on several examples. V současné době díky vysoce výkonným počítačům a zdokonaleným počítačovým periferiím, je možné vcelku snadno upravovat digitální fotografie. Moderní programy na zpracování obrazů umožňují vytořit padělky fotografií, které nejsou rozpoznatelné prostým okem. Jednou z nejčastějších technik takových úprav je převzorkování jedné fotografie a spojení s druhou, při tom je použita změna měřítka a případné otočení části obrazu. V této práci je navržena metoda, která automaticky detekuje převzorkovanou část obrazu. Výsledky a použitelnost metody je ukázána na řadě příkladů Keywords: image processing; image tampering; resampling Available at various institutes of the ASCR Periodic Properties Of Resampled Images

In recent years, due to the advent of high-performance commodity hardware and improved human computer interfaces, it has become relatively easy to create fake images. Modern, image processing ...

Mahdian, Babak; Saic, Stanislav
Ústav teorie informace a automatizace, 2007

On Divergences of Finite Measures and their Statistical Applicability
Stummer, W.; Vajda, Igor
2007 - English Keywords: divergences of finite measures; Pinsker's inequality; statistical censoring; Bayesian decisions Available at various institutes of the ASCR On Divergences of Finite Measures and their Statistical Applicability

Stummer, W.; Vajda, Igor
Ústav teorie informace a automatizace, 2007

Selection of Most Informative Variables in Statistical Pattern Recognition
Pudil, Pavel; Somol, Petr; Haindl, Michal
2007 - English The research report gives an overview of feature selection techniques in statistical pattern recognition with particular emphasis to methods developed by the researchers participating in MATEO Centre of Mechatronics project. Besides discussing the advances in methodology it attempts to put them into a taxonomical framework. The methods discussed include the latest variants of the optimal algorithms, enhanced sub-optimal techniques and the simultaneous semi-parametric probability density function modelling and feature space selection method. Some related issues are illustrated on real data by means of the Feature Selection Toolbox software. Výzkumná zpráva obsahuje přehled metod výběru příznaků ve statistickém rozpoznávání s důrazem na metody vyvinuté výzkumníky projektu MATEO Centre of Mechatronics. Diskutované metody zahrnují nejnovější verze optimalizačních algoritmů, sub-optimální techniky a modelování simultánní semi-parametrické pravděpodobnostní hustoty a metody výběru příznaků. Metody jsou ilustrované na reálných datech pomocí programu Feature Selection Toolbox. Keywords: feture selection Available at various institutes of the ASCR Selection of Most Informative Variables in Statistical Pattern Recognition

The research report gives an overview of feature selection techniques in statistical pattern recognition with particular emphasis to methods developed by the researchers participating in MATEO Centre ...

Pudil, Pavel; Somol, Petr; Haindl, Michal
Ústav teorie informace a automatizace, 2007

Application of Phi-divergence to Estimation in Continuous Families
Vajda, Igor
2007 - English Application of Phi-divergence to Estimation in Continuous Families Keywords: probabilistic measure; statistical estimation; phi-divergence Available at various institutes of the ASCR Application of Phi-divergence to Estimation in Continuous Families

Application of Phi-divergence to Estimation in Continuous Families

Vajda, Igor
Ústav teorie informace a automatizace, 2007

Analysis of Meteorological Images for Storm Detection in Short-term Weather Forecast
Vaníčková, P.; Šroubek, Filip
2007 - English Analysis of Meteorological Images for Storm Detection in Short-term Weather Forecast Keywords: Image processing; Weather forecasting; Storm detection Available at various institutes of the ASCR Analysis of Meteorological Images for Storm Detection in Short-term Weather Forecast

Analysis of Meteorological Images for Storm Detection in Short-term Weather Forecast

Vaníčková, P.; Šroubek, Filip
Ústav teorie informace a automatizace, 2007

Toolbox Jobcontrol with GUI. Program
Novák, Miroslav; Tesař, Ludvík
2007 - English Jobcontrol is a toolbox merging a user-friendly graphical user interface with a powerful set of utilities for system identification employing Gaussian mixture model. It is implemented as a set of M-scripts and MEX-binary executables for the Matlab computing environment. It suits to the goal of finding a suitable structure for the given data and then it creates the optimal controller by tuning its parameters to fit the user's requirements. Keywords: Controller design; System identification Available at various institutes of the ASCR Toolbox Jobcontrol with GUI. Program

Jobcontrol is a toolbox merging a user-friendly graphical user interface with a powerful set of utilities for system identification employing Gaussian mixture model. It is implemented as a set of ...

Novák, Miroslav; Tesař, Ludvík
Ústav teorie informace a automatizace, 2007

On Structure of Predictors Allowing Distributed Dynamic Bayesian Decision-Making
Šmídl, Václav
2007 - English Decentralized adaptive control is based on the use of many local controllers in parallel, each of them estimating its own local model and pursuing its local aims. If each controller designs its strategy using only its own model, the resulting control may be poor since consequences of actions of the neighbors are not taken into account. We seek a way how to improve algorithm of decision strategy design of a single local controller without significant increase in complexity of the local model or complexity of the design procedure. In this paper we study variants of distributed dynamic programming that could be evaluated locally. Specifically, we will investigate variants of the fully probabilistic control strategy design. Distributed and cen- tralized control strategies will be compared. Distribuované adaptivní řízení spočívá v paralelním běhu několika adaptivních regulátorů, z nichž každý odhaduje svůj model a navrhuje valstní strategii řízení. Pokud by takový regulátor nevzal do úvahy přítomnost ostatních regulátorů, může dojít k nežádoucímu chování. V této práci se zabýváme Bayesovským přístupem k problému. Cílem je zjistit jaká struktura centralizovaného prediktoru umožňuje návrh distribuované strategie. Keywords: distributed control; Bayesian decision-making; decentralized adaptive control; dynamic programming Available at various institutes of the ASCR On Structure of Predictors Allowing Distributed Dynamic Bayesian Decision-Making

Decentralized adaptive control is based on the use of many local controllers in parallel, each of them estimating its own local model and pursuing its local aims. If each controller designs its ...

Šmídl, Václav
Ústav teorie informace a automatizace, 2007