Unsupervised visual object categorization

The number of digital images has been increasing exponentially in the last few years. People have problems managing their image collections and finding a specific image. An automatic image categorization system could help them to manage images and find specific images. In this thesis, an unsupervised visual object categorization system was implemented to categorize a set of unknown images. The system is unsupervised, and hence, it does not need known images to train the system which needs to be manually obtained. Therefore, the number of possible categories and images can be huge. The system implemented in the thesis extracts local features from the images. These local features are used to build a codebook. The local features and the codebook are then used to generate a feature vector for an image. Images are categorized based on the feature vectors. The system is able to categorize any given set of images based on the visual appearance of the images. Images that have similar image regions are grouped together in the same category. Thus, for example, images which contain cars are assigned to the same cluster. The unsupervised visual object categorization system can be used in many situations, e.g., in an Internet search engine. The system can categorize images for a user, and the user can then easily find a specific type of image.

Comparison of Local Feature Detectors and Descriptors for Visual Object Categorization

Local features are used in many computer vision tasks including visual object categorization, content-based image retrieval and object recognition to mention a few. Local features are points, blobs or regions in images that are extracted using a local feature detector. To make use of extracted local features the localized interest points are described using a local feature descriptor. A descriptor histogram vector is a compact representation of an image and can be used for searching and matching images in databases. In this thesis the performance of local feature detectors and descriptors is evaluated for object class detection task. Features are extracted from image samples belonging to several object classes. Matching features are then searched using random image pairs of a same class. The goal of this thesis is to find out what are the best detector and descriptor methods for such task in terms of detector repeatability and descriptor matching rate.

Jäsenkategorisoinnin analyysi kulttuurisena metodina

Summary: Membership categorization device as a cultural method

Suomen implisiittisten refleksiivien semanttinen tulkinta

Summary: The semantic interpretation of Finnish implicit reflexives

Capturing Weak Signals for Foresight Building

Tulevaisuuden hahmottamisen merkitys heikkojen signaalien avulla on korostunut viime vuosien aikana merkittävästi,koska yrityksen liiketoimintaympäristössä tapahtuvia muutoksia on ollut yhä vaikeampaa ennustaa historian perusteella. Liiketoimintaympäristössä monien muutoksien merkkejä on ollut nähtävissä, mutta niitä on ollut vaikea havaita. Heikkoja signaaleja tunnistamalla ja keräämällä sekä reagoimalla tilanteeseen riittävän ajoissa, on mahdollista saavuttaa ylivoimaista kilpailuetua. Kirjallisuustutkimus keskittyy heikkojen signaalien tunnistamisen haasteisiin liiketoimintaympäristöstä, signaalien ja informaation kehittymiseen sekä informaation hallintaan organisaatiossa. Kiinnostus näihin perustuu tarpeeseen määritellä heikkojen signaalien tunnistamiseen vaadittava prosessi, jonka avulla heikot signaalit voidaan huomioida M-real Oyj:n päätöksenteossa. Kirjallisuustutkimus osoittaa selvästi sen, että heikkoja signaaleita on olemassa ja niitä pystytään tunnistamaan liiketoimintaympäristöstä. Signaaleja voidaan rikastuttaa yrityksessä olevalla tietämyksellä ja hyödyntää edelleen päätöksenteossa. Vertailtaessa sekä kirjallisuustutkimusta että empiiristä tutkimusta tuli ilmi selkeästi tiedon moninaisuus; määrä,laatu ja tiedonsaannin oikea-aikaisuus päätöksenteossa. Tutkimuksen aikana kehittyi prosessimalli tiedon suodattamiselle, luokittelulle ja heikkojen signaalien tunnistamiselle. Työn edetessä prosessimalli kehittyi osaksi tässä työssä kehitettyä kokonaisuutta 'Weak Signal Capturing' -työkalua. Monistamalla työkalua voidaan kerätä heikkoja signaaleja eri M-realin liiketoiminnan osa-alueilta. Tietoja systemaattisesti kokoamalla voidaan kartoittaa tulevaisuutta koko M-realille.

Exploration of the social dimension of corporate responsibility in a welfare state

This study focuses on corporate social responsibility (or CSR)as the latest dimension to emerge in the corporate responsibility and sustainability agenda, which in the recent past has rapidly risen to the top of the list of concerns for civil societies worldwide. Despite the continuing debates and discussions about the scope, benefits, and impacts of CSR to business and community in various sectors, levels, and types of society, many companies have moved forward to confront the opportunities and challenges of CSR. Thus, this study is about those proactive companies with a focus on the importance of CSR and its management inside and outside the company. It is an exploration and learning from the experience of Finnish companies, as well as other actors interested or involved in shaping the course of CSR, locally and globally. It also looks closely at how national culture affects the views, thinking, and management of CSR in a welfare state. This dissertation primarily draws on the analyses of information collected from a series of qualitative interviews and the existing literature in the area. This is complemented by an analysis of written and published documents on CSR from various sources. The results of the study give insightful information and detailed descriptions of a roadmap useful in learning and understanding CSR in Finnish companies. Despite the varying conceptual connotations, essential roadmap indicators point to the importance of framing CSR within the corporate responsibility concept, Finnish development and the welfare state system, globalization, stakeholders, and the pursuit of sustainable development as the main drivers of CSR, the remarkable progress of CSR in companies, and identification of key management areas and practices relevant to CSR. Similarly,the study reveals the importance of culture as essential in understanding and learning CSR. Finnish culture has a positive influence on the views, thinking, and management practices of CSR issues. Such a positive influence of culture, therefore, makes it easy for business people to discuss and understand CSR, because those CSR issues are already considered common and taken-for-granted by Finns and are implicit in the welfare state provisions. The experience of Finnish companies in implementing CSR policies in the supply chain is a concrete proactive step in advancing the message of CSR, that is, to bring companies and suppliers together to work on improving and strengthening relationships towards socially responsible practices worldwide. Such a forward step to deal with CSR issues in the supply chain reflects the companies' commitments and belief that CSR can be managed with the suppliers and gain positive benefits. Despite the problems and complexities, particularly in the global supply chain, managing CSR for Finnish companies presents new opportunities and challenges that are expected to intensify in the near future. The focus on CSR policy implementation inthe supply chain points to the importance of companies taking initiatives and forging cooperation with suppliers with the aim of addressing and improving CSR questions in the supply chains. The proactive stance of Finnish companies toward CSR is complemented by the active supporting role of important societalactors such as the government and NGOs. These actors carry out various promotional efforts and campaigns, thus bringing CSR into the mainstream of Finnish companies and strengthening the synergistic learning about CSR within the Finnish business and civil circles. The efforts of the government and NGOs to promote CSR are indicative of the importance of multipartite involvement and the emergence of better civil regulations. Likewise, their drive to learn from each other, exchange experiences, and contribute in CSR debates facilitated the evolution of CSRnetworks in the country. The results of this study add to the mounting evidence that CSR, in general, has created a new dimension in managing corporate sustainability. This study provides compelling empirical evidence and some direct quotations about CSR in the Finnish context. This information can be used to learn and gain new useful insights, approaches, and concepts for managing CSR.

Supervised localimage feature detection

This thesis is about detection of local image features. The research topic belongs to the wider area of object detection, which is a machine vision and pattern recognition problem where an object must be detected (located) in an image. State-of-the-art object detection methods often divide the problem into separate interest point detection and local image description steps, but in this thesis a different technique is used, leading to higher quality image features which enable more precise localization. Instead of using interest point detection the landmark positions are marked manually. Therefore, the quality of the image features is not limited by the interest point detection phase and the learning of image features is simplified. The approach combines both interest point detection and local description into one phase for detection. Computational efficiency of the descriptor is therefore important, leaving out many of the commonly used descriptors as unsuitably heavy. Multiresolution Gabor features has been the main descriptor in this thesis and improving their efficiency is a significant part. Actual image features are formed from descriptors by using a classifierwhich can then recognize similar looking patches in new images. The main classifier is based on Gaussian mixture models. Classifiers are used in one-class classifier configuration where there are only positive training samples without explicit background class. The local image feature detection method has been tested with two freely available face detection databases and a proprietary license plate database. The localization performance was very good in these experiments. Other applications applying the same under-lying techniques are also presented, including object categorization and fault detection.

Sliding Mode Control for Switched Mode Power Supplies

Control applications of switched mode power supplies have been widely investigated. The main objective ofresearch and development (R&D) in this field is always to find the most suitable control method to be implemented in various DC/DC converter topologies. Inother words, the goal is to select a control method capable of improving the efficiency of the converter, reducing the effect of disturbances (line and load variation), lessening the effect of EMI (electro magnetic interference), and beingless effected by component variation. The main objective of this research work is to study different control methods implemented in switched mode power supplies namely (PID control, hysteresis control, adaptive control, current programmed control, variable structure control (VSC), and sliding mode control (SMC). The advantages and drawbacks of each control method are given. Two control methods, the PID and the SMC are selected and their effects on DC/DC (Buck, Boost, and Buck-Boost) converters are examined. Matlab/SimulinkTM is used to implement PID control method in DC/DC Buck converter and SMC in DC/DC (Buck, and Buck Boost) converters. For the prototype, operational amplifiers (op-amps) are used to implement PID control in DC/DC Buck converter. For SMC op-amps are implemented in DC/DC Buck converter and dSPACETM is used to control the DC/DC Buck-Boost converter. The SMC can be applied to the DC/DC (Buck, Boost, and Buck-Boost) converters. A comparison of the effects of the PID control and the SMC on the DC/DC Buck converter response in steady state, under line variations, load variations, and different component variations is performed. Also the Conducted RF-Emissions between the PID and SMC DC/DC Buck Converter are compared. The thesis shows that, in comparison with the PID control, the SMC provides better steady-state response, better dynamic response, less EMI, inherent order reduction, robustness against system uncertainty disturbances, and an implicit stability proof. Giving a better steady-state and dynamic response, the SMC is implemented in a DC/DC resonant converter. The half-wave zero current switching (HWZCS) DC/DC Buck converter is selected as a converter topology. A general guideline to select the tank component values, needed for the designing of a HWZCS DC/DC Buck, is obtained. The implementation of the SMC to a HWZCS DC/DC Buck converter is analysed. The converter response is investigated in the steady-state region and in the dynamic region.

Perusterveydenhuollon ja erikoissairaanhoidon käytön seuranta ja ennakointi kuntanäkökulmasta

Follow-up of utilisation and prediction of primary health care and hospital care from the municipality point of view. Planning, follow-up, and evaluation of primary health care within municipality entail comprehensive information about factors that influence health. In addition to populationbased research, various statistical data and registries serve as sources of information. The present study examined utilisation of primary health care and hospital care with the existing databases, registries, and categorization of Diagnosis Related Groups (DRGs) from the municipality (purchaser) point of view. Research involving the cases of Paimio, Sauvo, and Turku as examples of municipalities pointed out that, even in the small municipalities, it is possible to assess and predict health services to be offered to the inhabitants by following databases and registries. Health-related databases and registries include a plenty of possible uses that have not adequately been employed at the level of municipality. Descriptive futures research and community analysis formed the framework of the study. Descriptive futures research may be used to establish predictions based on past developmental traditions, and quantitative time trend analyses may be employed to make estimations about future events. Community analysis will assist in making conclusions about population- based health care needs, in assessing the functionality or effectiveness of the health care system, and in appropriately targeting limited resources. The aim of the present study was to describe the health service profile so that the arrangements and planning of health services as well as the contract negotiations of hospital care become easier within municipalities. Another aim was to assess the application of Hilmo (registry for posting hospital care periods), Aitta and Sotka (statistical databases) for the purposes of resource planning in the procurement of hospital care. A third aim was to evaluate how the system of the DRGs adapts in the prediction of retaining health services within short (1-year), intermediate (5-year) and long range (10-15-year) intervals. The findings indicated that the follow-up of primary health care utilisation combined with follow-up of hospital care utilisation allows municipalities to plan and predict health services when databases are applied. Information about the past contacts with the databases has indicated that the health care culture and incidence of disease change rather slowly in the area of investigation. For the purposes of health care research, it is recommended that methods of application used in making predictions about health care utilisation need to be further developed

Intelligent Internet Data Searching

In this thesis author approaches the problem of automated text classification, which is one of basic tasks for building Intelligent Internet Search Agent. The work discusses various approaches to solving sub-problems of automated text classification, such as feature extraction and machine learning on text sources. Author also describes her own multiword approach to feature extraction and pres-ents the results of testing this approach using linear discriminant analysis based classifier, and classifier combining unsupervised learning for etalon extraction with supervised learning using common backpropagation algorithm for multilevel perceptron.

Implisiittisen kalibroinnin soveltamismahdollisuudet kemiantekniikassa

Työn taustalla oli tavoite parantaa erään teollisuusprosessin toimintaa ja sen tuottoa mallintamalla reaktiovaiheen alussa tapahtuvan välituotteen muodostumisen reaktiokinetiikka sekä perinteisellä tavalla että implisiittisellä kalibroinnilla. Toisena tavoitteena oli selvittää, kuinka implisiittistä kalibrointia voidaan yleisemmin hyödyntää kemiantekniikassa. Implisiittinen kalibrointi on menetelmä, jolla voidaan ratkaista jonkin teoreettisen mallin parametrit suoraan epäsuorasta mittausdatasta (esimerkiksi spektreistä) lähes kokonaan ilman off-line analyysejä. Tämän työn kirjallisuusosassa on esitetty implisiittisen kalibroinnin toimintaperiaate sekä lyhyesti FTIR-spektrometrian perusteita. Työn kokeellisessa osassa on estimoitu tutkitun välituotteen muodostumisen kineettiset parametrit sekä tavanomaisella parametriestimoinnilla että implisiittisellä kalibroinnilla. Lisäksi kokeellisessa osassa on selvitetty lyhyesti tutkitun prosessin FTIR-spektrien lämpötilariippuvuuksia ja esitetty neljä mahdollista uutta sovelluskohdetta implisiittiselle kalibroinnille. Tavanomaisella parametriestimoinnilla saatiin estimoitua varsin yksiselitteiset arvot kineettisille parametreille. Myös mallin sovitus koedataan on hyvä kolmessa kokeessa viidestä. Parametriestimointi implisiittisellä kalibroinnilla onnistui lupaavasti vaikka tulokset eivät ole aivan niin hyviä kuin tavanomaisessa parametriestimoinnissa. Parhaat tulokset implisiittisessä kalibroinnissa saavutettiin suoralla kalibrointitavalla GRR (Generalized Ridge Regression)-kalibrointimenetelmää käyttämällä.

Cost allocation of corporate IT-services

Cost allocation is an inescapable problem in nearly every organization and in nearly every facet of accounting. Within large corporations there are several different types of units, like profit-making business units and non-profit service units. In order to evaluate the performance of the business units and to fund the operations of service units, the expenses of service production need to be allocated to the business units benefiting from the services.The objective of this thesis was to find good and fair allocating factors for the costs of corporate wide IT services. In order to reach this objective, the cost allocation process was studied in general and an overview of cost structure was established. All possible cost driver candidates were mapped and their good and bad properties were weighed. The cost allocation problem was handled separately according to organizational division of corporate IT department: infrastructure, administrative systems, sales system and e-business. The emphasis was on two largest cost groups: infrastructure costs and sales system costs. As a result of the study an allocation model is presented. It contains categorization of the costs, selected cost drivers and cost distributions for the current year.