WWW.DISSERTATION.XLIBX.INFO
FREE ELECTRONIC LIBRARY - Dissertations, online materials
 
<< HOME
CONTACTS



Pages:     | 1 |   ...   | 2 | 3 || 5 | 6 |   ...   | 8 |

«EXTRACTION OF CONTEXTUAL KNOWLEDGE AND AMBIGUITY HANDLING FOR ONTOLOGY IN VIRTUAL ENVIRONMENT A Dissertation by HYUN SOO LEE Submitted to the Office ...»

-- [ Page 4 ] --

Definition 5. Mathematical definition of virtual ontology

Virtual ontology is defined as a graph:

–  –  –

In definition 5, VC is a set consisting of a virtual space layer in the Metaearth architecture. C 1 and C 2 are the components consisting of the library layer and ontology layer. Finally, the context C 2 is mapped to C 1 using CR and C 1 is linked to each VC through MR. From this viewpoint, MR can be considered as a component of the mapping layer in Metaearth architecture. Virtual ontology can be represented using this Metaearth architecture. The next sections explain how we can automatically or semiautomatically generate Metaearth architecture from scenes.

4. CONSTRUCTION OF VIRTUAL ONTOLOGY USING ONE SCENE

–  –  –

This section describes how to generate Metaearth architecture with virtual ontology from a 2D image. The process starts by over-segmenting a scene (the 2D image) using the proposed algorithm. Recall that the suggested approach generates relationships between components which are not adjacent as an additional advantage.

Figure 16. A general procedure of existing scene understanding algorithm.

Next, we group the segmented regions via the object merging process and the semantic merging process. Finally, the 2D scene is converted into Metaearth architecture (described in Section 3.2). Figure 16 shows the general procedures of existing scene understanding techniques using IDEF 0 diagram and Figure 17 illustrates the procedure broken down by steps and sub-steps.

Figure 17. The general procedures for generating virtual ontology from a scene.

The three sub-steps depicted in Figure 17 and detailed in Figure 18 are: 1) fuzzy color-based over-segmentation; 2) object merging; and 3) semantic merging.

1) Fuzzy color-based over-segmentation The approach considers over-segmentation as a pre-processor which is dependent on prior knowledge or training mechanisms. Most methods such as Chung and Fung [58] use a pre-defined fuzzy color for segmentation, but the result and quality are not stable, because uniform color quantization or trained fuzzy color are not extracted from an input.

Figure 18. Detailed sub-processes of virtual ontology generation from a scene.

2) Object merging The segmentation process generates an initial graph structure using adjacency among segments. This graph forms our initial layout for generating the virtual space layer in Metaearth architecture.

3) Semantic merging Having constructed the virtual space layer, we can now generate the core layer and mapping layer via the semantic merging process.

The following sections explain the three sub-processes with suitable examples and implementations.

4.2 FUZZY COLOR-BASED OVER-SEGMENTATION AND GENERATION OF

–  –  –

Over-segmentation is considered as a preprocessing stage. Using an image with 1024 768 pixels, the initial number of overall pixels is 786,432( 1024  768). Note that handling around 700,000 variables is considered a large-scale optimization issue. In addition, each pixel has many properties such as true color, hue or lamination. In true color, a pixel has red (R), green (G) and blue (B). Each R, G, B value has one value out of 0 to 255 (  28 ). If a pixel has a true color, the size of the variable increases to 786,432  2 24. Obviously, the increased size contributes to computation complexity;

thus, in order to reduce the variables we conduct random sampling to select a certain amount of pixels. Considering the input image’s height and width, we extract 1/ 700 pixels. Figure 19 shows the randomly sampled pixels.

–  –  –

The true colors in the extracted pixels are selected by 1/700 scales of our original image size. Figure 20 shows the R, G, B plotting from the extracted pixels.

Figure 20. Red, Green and Blue color plotting from randomly extracted pixels.

The objective of this process is to obtain K-means fuzzy color quantization from the input image. The problem is to determine the K value. In most K-means algorithms or KNN approaches, the value K is assumed or determined using learning methods.

Figure 21 shows the general K-means algorithm.

–  –  –

To determine an effective K value, we modify Ray and Turi’s method [59].

Their method uses the intra-cluster distance measure and inner-cluster distance measure.

In their method, K value is determined by some condition such as minimizing the innercluster scatter and maximizing the intra-cluster separation.

Definition 6. Inner distance

–  –  –

the desirable K. In this manner, optimal K is calculated using a nonlinear programming in Table 5.

Table 5. K-fuzzy color determination using nonlinear programming.

–  –  –

The objective function of the nonlinear programming has four types of constraints. The first and second guarantee a locally minimizing ratio distance and the

–  –  –





distance between two clusters is usually larger and can be a trivial solution. The nonlinear programming allows us to calculate an optimal K value from the input image.

Figure 22 shows this process. Thus, we obtain the value 12 as the optimal K minimizing the ratio distance.

Figure 22. Determination of K using nonlinear programming.

Finally, the R, G, B color values from the randomly extracted pixels (Figure 20) are segmented into 12 regions (Figure 23).

Figure 23. Segmented regions using nonlinear programming.

After segmenting color spaces are acquired using an input source, fuzzy colors are extracted from each segmented region. As this fuzzy color extraction is based on the input source, it is called “dynamic fuzzy color extraction”. Other methods such as Chung and Fung [58] and Konstantinidis et al. [60] use a predefined fuzzy color or trained parameters for extracting fuzzy colors. In this concept, existing methods can be considered as “static fuzzy color generations”. In this dissertation, a triangular fuzzy membership function [61] is used for defining fuzzy color. Definition 9 represents triangular based fuzzy color.

Definition 9. Fuzzy color

–  –  –

2. Input image-based fuzzy color generation has superiority.

The first advantage means that a specific pixel can belong to many segments (one ~ to many mapping) with I i, j utility values. This characteristic is useful for determining ordering the next best segment in a true color-based color.

–  –  –

Finally, the image is re-segmented using the generated fuzzy colors. Figure 25 shows that each red boundary-based segment is obtained by measuring the distance between a pixel’s color and the generated fuzzy colors. We select the closed fuzzy color as an alternating color of the pixel. If the 8 adjacent pixels near the pixel have the same fuzzy color, it is determined as the same group.

–  –  –

However, fuzzy color-based segmented regions exhibit several problems, one of which is over-abstraction. Even though fuzzy color can map one pixel to multiple colors, the existence of the small fuzzy colors (= K fuzzy small colors) occurs in overabstraction, i.e. comparatively distinct colors are grouped in one segment. Figure 26 shows an example of over-abstraction.

–  –  –

From the 557 initially segmented regions, we select the first and second regions.

The second region as shown in Figure 26 (b), represents part of the cloud. It can be considered a substantially over-segmented region because it has only one context (cloud). The first region with three contexts, sky, cloud and building wall, is considered “over-abstraction”. The occurrence of over-abstraction due to the small number of fuzzy colors requires re-segmenting some regions. To prevent over-abstraction, the “cell division using EM algorithm”, a statistical method which is guaranteed to converge to the maximum likelihood estimator (MLE) for estimating a parameter, is proposed. An abbreviation of the Expectation-Maximization algorithm, EM algorithm is based on the concept of replacing one difficult likelihood maximization with a sequence of easier maximization whose limit is the answer to the original problem [62]. Table 6 delineates the procedure.

–  –  –

Step 1 : In each fuzzy color segmented region, R, G, B histogram is generated.

Step 2 : Estimation of Gaussian mixture distributions In each R,G,B histogram, a Gaussian mixture distributions is mapped.

–  –  –

Step 2.1 : An estimation of unimodal Gaussian mixture distribution using direct fitting - regression Step 2.

2 : An estimation of multimodal Gaussian mixture distribution using EM algorithm

–  –  –

Here, we assume that each color histogram from a specific segment can be explained using Gaussian mixture distributions. Using the suggested algorithm, the overabstracted region is divided into regions. Figure 27 shows the result.

–  –  –

In this figure, one region is divided into 4 regions. The histogram of the red and blue color are fitted by a bimodal Gaussian mixed distribution and the histogram of the green color is fitted by a unimodal Gaussian mixed distribution. Finally, the 4 inner regions ( 4  2 1 2 ) are segmented using cell division.

Another problem with the suggested fuzzy color-based segmentation is noise.

Figure 25 shows that our segmented image comprises many small areas, or “noise regions”. However, even though each has a specific color and context, we can safely ignore in the interests of over-segmentation. We use 1,000 pixels as a predefined size.

Figure 28 shows the maximum size of the noise region (the blue box).

–  –  –

This noise region is merged with other noise regions or non-noise regions. As candidate cells to be merged with a noise region, adjacent regions are listed and fuzzy color intensities in each R,G,B domain are compared. If their differences are less than our pre-defined small value, we merge the noise cell with the targeted candidate cell. In this process, the threshold values are different in each R, G, B color domain. Each threshold value is determined by the estimated parameters of the Gaussian mixture distributions as shown in Table 5. From the means and variance estimator in each R,G,B domain, we calculate the range i, j   i, j, where i represents each color domain ( 1  R, 2  G and 3  B ) and j means j th as j th estimator. The reason to use 1  is that it is enough for extracting the threshold values empirically. From each i, j   i, j plotting, the minimum inter-distance is calculated and used for a threshold value. Figure 29 is a graphic illustration.

Figure 29. Extraction of a threshold value for noise region handling in the Red color

–  –  –

If the minimum value is 0 or negative value while extracting  i;i 1, 2,3, the next positive value is determined as a  i;i 1, 2,3. After extracting  i;i 1, 2,3, the R,G,B intensity differences between the target noise region and adjacent regions are measured. If these intensities are less than  i;i 1, 2,3, two regions are determined to be merged. Then we merge the region with smaller pixels into the pixel with larger pixels. Figure 30 shows the result after cell division and noise region handling.

–  –  –

Compared with the initial segmentations (Figure 25), our finalized image of 78 regions lacks noise and many over-abstracted regions.

The final step consists of constructing a virtual space layer for Metaearth architecture.

–  –  –

Figure 31 shows the fuzzy color-based over-segmented region after mapping its fuzzy colors. In this example our 1024 * 768 image is compressed into 78 regions.

–  –  –

Our objective is the construction of Metaearth architecture containing virtual ontology. The following section describes the first two steps shown in Table 4: the identification of virtual components from the real world and the generation of virtual space layers. In a general 2D image, we replace the first step with the fuzzy color-based segmentation so that each region represents one virtual component. This proposition is supported by the inference that each virtual component has unique color and the full virtual model can be assembled using these components. Thus, the over-segmented regions are converted into a graph structure using each region’s bonding intensity ( BI i, j )

defined as:

Definition 10. Bonding intensity ( BI i, j ) The bonding intensity ( BI i, j ) is the number of shared pixels between i th region and j th region.

Proposition 1. The symmetry of the bonding intensity

–  –  –

The proof of Proposition 1 is trivial. For the adjacency graph, the value in the i th row and the j th column is BI i, j ( BI j,i ). After constructing the adjacency graph, we construct the initial graph in which each vertex represents each region. The information for a vertex and an edge appears in Figure 32.

–  –  –

In this figure, the Z depth and texture information can be skipped by the input sources.

For example, the initial graph using Section 5’s methods has Z-depth information.

Figure 33 shows the generated initial graph from fuzzy color-based over segmented region.

Figure 33. Conversion from fuzzy color-based segments into a graph.

–  –  –

This generated graph becomes the virtual space layer in Metaearth architecture.



Pages:     | 1 |   ...   | 2 | 3 || 5 | 6 |   ...   | 8 |


Similar works:

«Doctoral theses at NTNU, 2015:289 Doctoral thesis Stine Bruland Stine Bruland UNDERNEATH THE MARGOSA TREE Re-creating meaning in a Tamil family after war and migration ISBN 978-82-326-1234-5 (printed ver.) ISBN 978-82-326-1235-2 (electronic ver.) ISSN 1503-8181 Doctoral theses at NTNU, 2015:289 NTNU Norwegian University of Science and Technology Thesis for the degree of Philosophiae Doctor Faculty of Social Sciences and Technology Management Department of Social Anthropology Stine Bruland...»

«INVESTIGATION OF OPTICAL LOSS CHANGES IN SILOXANE POLYMER WAVEGUIDES DURING THERMAL CURING AND AGING A Dissertation Presented to The Academic Faculty by Shashikant G. Hegde In Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the G.W. Woodruff School of Mechanical Engineering Georgia Institute of Technology April 2008 COPYRIGHT 2008 BY SHASHIKANT G. HEGDE INVESTIGATION OF OPTICAL LOSS CHANGES IN SILOXANE POLYMER WAVEGUIDES DURING THERMAL CURING AND AGING Approved...»

«Multi-Party Electronic Payments for Mobile Communications Michael Peirce A thesis submitted for the degree of Doctor of Philosophy in Computer Science University of Dublin, Trinity College Department of Computer Science October 31st, 2000 Declaration I hereby declare that: (a) This thesis has not been submitted as an exercise for a degree at this or any other University. (b) This thesis is entirely the work of the author, except where otherwise stated. (c) The Trinity College Library may lend...»

«WHO PUT THE FUN IN FUNCTIONAL? FUN AT WORK AND ITS EFFECTS ON JOB PERFORMANCE By ERIN RAE FLUEGGE A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008 1 © 2008 Erin Rae Fluegge 2 To Mom & Dad 3 ACKNOWLEDGMENTS First and foremost, I thank my fantastic husband, Rob Woolf, for supporting my education habit throughout the whole nine years we have been together....»

«LIBERTY UNIVERSITY TO THE JEW FIRST: A SOCIO-HISTORICAL AND BIBLICAL-THEOLOGICAL ANALYSIS OF THE PAULINE TEACHING OF ‘ELECTION’ IN LIGHT OF SECOND TEMPLE JEWISH PATTERNS OF THOUGHT A DISSERTATION SUBMITTED TO THE FACULTY OF LIBERTY BAPTIST THEOLOGICAL SEMINARY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY BY ANTHONY CHADWICK THORNHILL LYNCHBURG, VIRGINIA JUNE 2013 Copyright 2013 A. Chadwick Thornhill All Rights Reserved ii APPROVAL SHEET TO THE JEW FIRST:...»

«The Pennsylvania State University The Graduate School Department of Anthropology SETTLEMENT AND POPULATION AT PIEDRAS NEGRAS, GUATEMALA A Thesis in Anthropology by Zachary Nathan Nelson © 2005 Zachary Nathan Nelson Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy May 2005 Abstract My dissertation examines the relationship between settlement and population at Piedras Negras, Guatemala. This Classic Maya center developed from a small village into a...»

«Learning Distance Functions for Exemplar-Based Object Recognition by Andrea Lynn Frome B.S. (Mary Washington College) 1996 A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Computer Science in the GRADUATE DIVISION of the UNIVERSITY OF CALIFORNIA, BERKELEY Committee in charge: Professor Jitendra Malik, Chair Professor Michael I. Jordan Professor Bruno Olshausen Fall 2007 The dissertation of Andrea Lynn Frome is approved: Professor...»

«UNIVERSITETET I BERGEN Coupled surface process and tectonic modelling of extension-inversion tectonics in the Pyrenees Dissertation for the degree of Philosophiae Doctor (PhD) Zoltán Erdős University of Bergen 2014 Preface The research presented in this thesis was conducted at the Department of Earth Science, University of Bergen (Norway), and in parallel at the Institute of Earth Sciences, Joseph Fourier University (Grenoble, France) under the supervision of Prof. Ritske Huismans, Prof....»

«Understanding Civic Engagement among Youth in Diverse Contexts By Holly Lynn Karakos Dissertation Submitted to the faculty of the Graduate School of Vanderbilt University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY in Community Research and Action May, 2015 Nashville, Tennessee Approved: Maury Nation, Ph.D. Andrew J. Finch, Ph.D. Paul Speer, Ph.D. Sonya Sterba, Ph.D. ABSTRACT Research suggests that youth civic engagement is beneficial for individuals and...»

«THE SOFTWARE SERVICE BAY: A KNOWLEDGE BASED SOFTWARE MAINTENANCE METHODOLOGY by JONATHAN IRVINE MALETIC DISSERTATION Submitted to the Graduate School of Wayne State University, Detroit, Michigan in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY MAJOR: COMPUTER SCIENCE Approved by: _ Adviser Date _ _ _ Acknowledgements. I thank my advisor and mentor, Robert G. Reynolds. All my work is due to his help and direction. I would also like to thank my committee, William...»

«IDENTIFYING HUMAN AND ANIMAL FAECAL CONTAMINATION IN SURFACE AND DRINKING WATER: NEW CONCEPTS AND APPROACHES Inauguraldissertation zur Erlangung der Würde eines Doktors der Philosophie vorgelegt der Philosophisch-Naturwissenschaftlichen Fakultät der Universität Basel von Melanie Wicki aus Basel Basel, 2011 Originaldokument gespeichert auf dem Dokumentenserver der Universität Basel edoc.unibas.ch Dieses Werk ist unter dem Vertrag „Creative Commons Namensnennung-Keine kommerzielle...»

«Reconstructing Student Conceptions of Climate Change; An Inquiry Approach A DISSERTATION SUBMITTED TO THE FACULTY OF UNIVERSITY OF MINNESOTA BY J Collin McClelland IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY Gillian H. Roehrig, Advisor August 2015 © J Collin McClelland 2015 Acknowledgements The journey through my doctoral program was long and arduous, accompanied by growth beyond my expectations. Sustaining me through this journey and adding to my personal...»





 
<<  HOME   |    CONTACTS
2016 www.dissertation.xlibx.info - Dissertations, online materials

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.