Jaroslav Zacek

Detailed Analysis of REA Ontology

The paper addresses REA (Resource-Event-Agent) domain specific ontology that is primarily focused on value modeling in business processes. REA ontology which historically originates from accounting information systems, gradually developed to cover all areas where value modeling can be utilized. After a short introduction, a core REA pattern is introduced and analyzed from the view of its basic entities and the relationships between them. Next, additional crucial concepts of REA ontology and their relationships are gradually elucidated and analyzed from the view of DEMO (Design & Engineering Methodology for Organizations). The paper also describes the current definition of economic transaction that is used as a basis for REA state machine. The discussion and conclusion sections summarize and assess the pros and cons of REA ontology and propose a way forward for further research.

CEM: Class executing modelling

Abstract This paper analysis approaches and possibilities of executive model aimed to MDA approach. The second part of the article proposes guideline to create executive model and describes basic interactions to object oriented approach. Annotations have been used for executive model object extension. Reflection concept has been used for model execution. Proposed model supports new type of extended object with enhanced metadata model as well as regular objects with no additional metadata description. According to use object with no additional description the model supports third-part components and supports reusability. The model will be applied to LFLC package developed by Institute for Research and Applications of Fuzzy Modeling, University of Ostrava.

Machine learning approach to point localization system

The article introduces point localization systems in 3D Euclidean space based on neural networks. There are two models presented. The first one identified distances between a randomly generated point and a reference points in the defined domain. Then a neural network uses the obtained distances as its inputs to determine the actual position of the point in the domain space. Due to a relatively good accuracy that was obtained during the experimental study, the proposed model based on neural networks was used in the second model as an acoustic Motion Capturing system (MoCap). MoCap system is represented by a neural network that uses obtained distances between transmitters and a receiver as its inputs to determine an actual position of the receiver in space. We also propose a new way to minimize a training set by using ANFIS approach in this specific problem. All obtained results are summarized in the conclusion.

Applying ETL fuzzy filter for data warehouse size minimization

The paper proposes a new approach of data warehouse minimization by fuzzy-based ETL filter for ETL processes in business intelligence (BI) systems and discusses possibilities to tune up the process. First part introduces common company systems and possible data sources in the company. Second part states the problem with interpreting information in BI systems and explains a data representation in the BI systems. Third part of the paper defines a rule base and input and output values of the expert system. Last part of the paper proposes a two ways to minimize a data - modification border of the fuzzy set and omitting useless combinations of the linguistic variables and modifiers.

Modeling the Value Chain with Object-Valued Petri Nets

A substantial part of the economic theories is based on conversion and exchange process. These processes can be arranged in a value chain, which can be considered as a cyclic model with complex attributes. There is a serious problem how to express resources and their conversions in a complex cyclic model during the simulation and how to identify these converted resources in every step of the simulation. This paper introduces the Object- valued Petri (OV-PN) modification as a new formalism to create a cyclic model of the value chain. According to the modification we had to define a new path and pass of the OV-PN. We also had to determine new properties. Properties are based on the OV-PN and reflect needs of model requirements. A new formalism is verified on a common enterprise value chain.

Reusable Object-Oriented Model

This paper analysis approaches and possibilities of executive model aimed to MDA approach. The second part of the article proposes guideline to create executive model, describes basic interactions to object oriented approach and shows possibilities of creating a core of executable model in Java programming language. Annotations are used for executive model object extension. Reflection concept is used for model execution and synchronization provides extended Petri net formalism defined in [1]. The model has been tested on LFLC software package developed by IRAFM, University of Ostrava to prove the whole concept.

Training Set Fuzzification Towards Prediction Improvement

This article presents a method of fuzzification of variables using a histogram. This approach is used when creating an output vector of a training set that forms linguistic variables. An appropriate transformation of an input vector of the training sets was also proposed. Both of the aforesaid procedures were described in detail in the article. An extensive comparative experimental study with the following outcomes was carried out. The neural net which was adapted by the transformed training set showed a significantly better prediction than a neural network which was adapted by a training set without making any changes. The results of this experimental study were analyzed in the conclusion.

Modular neural network in the cloud

The paper proposes a new approach to implement common neural network algorithms in the cloud. First part of the paper defines the context and describes a usage of cloud computing concept in the robotics. Next part describes layers of the typical cloud service and shows the connection between cloud and legacy robot. Then we introduce a modular neural network concept and explain the topology of the neural network and connection between the physical and the logical architecture. Following part proposes a new architecture of the cloud service that implements a neural network algorithm. Last part shows a case study migration of the typical neural network topology into the cloud with respect to logical and physical topology.

Application of Linguistic Fuzzy-Logic Control in fast and sensitive technological processes

This paper presents the use of modern numerical methods such as Fuzzy Logic Control for control of fast and sensitive non-linear technological processes with sampling period 0.01 [s] or less. The paper presents a real application of the Linguistic Fuzzy-Logic Control, developed at the University of Ostrava for the control of physical models in the Intelligent Systems Laboratory. The paper presents an example of a sensitive non-linear model, such as a helicopter model and obtained results which show how modern information technologies can help to solve actual technical problems. The paper shows how the used technology can help people easily describe the control strategy. A special method based on the LFLC controller with partial components is presented in this paper followed by the method of automatic context change, which is very helpful to achieve more accurate control results. This technology and real models are used as a background for problem-oriented teaching, realized at the department for master students and their collaborative as well as individual final projects.

Neural Network Based Complex Visual Information Processing: Face Detection and Recognition

This paper focuses on the issue of detecting and recognizing faces. The work is divided into three main categories. The first part is about detection of faces in constrained conditions. The second part focuses on creation of a different recognition approach. The third one is about the test with robotic devices. However mobile devices (such as robots, small CCD cameras or cheaper cell phones) have many limitations i.e. images quality or very limited computing performance. With respect to limitations the system manages two substantial parts. The first one is responsible for detecting a face in an image. The second one is responsible for calculating the information featured in a face image and recognition of that information. The system is able to process faces in realtime with minimal computation performance and to use minimal space for storing its data. The proposed system was tested on a face database. We have used a FDDB benchmark for an exact comparison. 1 Face Detection and Face Recognition Methods In the early development of face detection [3], geometric facial features such as eyes, nose, mouth, and chins were explicitly used. Properties of the features and relations among them (e.g. areas, distances, angles) were used as descriptors for face recognition. Statistical learning methods are the mainstream approaches that have been used in building face recognition systems. Effective features are learned from training data and involve prior knowledge about faces. The appearance-based approaches, such as Principal Component Analysis (PCA) [11] or Linear Discriminant Analysis (LDA) [2], have significantly advanced the face recognition process. These approaches operate directly on an image-based representation (i.e. fields of pixel intensities). It extracts features in a subspace derived from training images. The most successful approach, so far, to handle the non-convex face segmentation works with local appearance-based features. These features are extracted using appropriate image filters. An advantage lies in a distribution of face images through local feature space, which is less affected by changes in facial appearance. Early works in this area include local features analysis (LFA) [10] and Gabor wavelet-based features [14]. Current methods are based on local binary pattern (LBP) [1]. There are many variants to the basic approaches like: Ordinal Features [8], Scale-Invariant Feature Transform (SIFT) [9], and Histogram of Oriented Gradients (HOG) [4]. While these features are general purpose, face specific local filters are learned from images [13]. Zacek V., Volná E. and Zacek J.. Neural Network Based Complex Visual Information Processing: Face Detection and Recognition. DOI: 10.5220/0005126800530060 In Proceedings of the International Workshop on Artificial Neural Networks and Intelligent Information Processing (ANNIIP-2014), pages 53-60 ISBN: 978-989-758-041-3 Copyright c 2014 SCITEPRESS (Science and Technology Publications, Lda.) The most famous early example of a face recognition system is due to Kohonen [7], who demonstrated the strength of a simple neural net that was able to perform face recognition for aligned and normalized face images. A face itself can be represented by a lot smaller number of eigenvectors. Nowadays eigenfaces are based on the work of Sirovich and Kirby [6] and use principal of component analysis. They start by creation of a feature space from a training set of all faces. Using the created feature space, the algorithm calculates additional information in the form of weights. Each face will have its weights, which are projected to the feature space for reconstruction of the face. The eigenfaces method expects that under the best-idealized conditions the variations between faces lie in a linear subspace. This means that classes are linearly separable. The reason behind the “upgrade” of eigenfaces approach is that it does not use class specific linear methods for dimension reduction. An example of a class specific method is Fisher’s Linear Discriminant (FLD) (fisherfaces) [1]. Another approach, Local Binary Patterns (LBP), uses the operator for description of the area surrounding the pixel. If the simplest algorithm is implemented it considerate the surrounding of 3x3 only. When the LBP code for an image is calculated, the edges are ignored for the lack of information. To gain more information, overlapping of the operators is used to obtain high and low frequency information about the neighbors. This means that the required space for storing the basic information about neighboring pixels would be too demanding. We have to use the data reduction to minimize the space requirements. Therefore we introduce uniform patterns. 2 Proposed Face Recognition System All systems described above have very good results by applying them to the image captured in standard conditions. However mobile devices (such as robots, small CCD cameras or cheaper cell phones) have many limitations i.e. images quality or very limited computing performance. The proposed system should be able to detect and recognize faces in the image despite the limitations. At first, we have to define a limitation conditions for the group of mobile devices [16]:  The image quality of the input device,  Real-time detection,  Processing speed requirements,  Memory requirements. With respect to limitations defined above the system has to state two substantial parts.The first part is responsible for detecting a face in an image. The second part is responsible for calculating the information featured in the fac *e image and recognition of that information. The system itself is proposed to be light-weighted. This means that the system should be able to process faces in real-time with minimal computation performance and to use minimal space for storing its data.