Textfeld: Copyright
Mag. Andrew Lindley
Leitermayergasse 31/10
A-1180 Wien
Tel: +43 650 3642799

MPEG-7 Feature Extraction Plugin

Download Master Thesis (pdf)                                                                                                                   

Textfeld: SOFTWARE:
Features XM Server:
standalone XM component
Java Functionality Wrapper
Multiuser Server Application
SOAP Binary File Handling (SwA)

Features Client Plugin:
MPEG-7 data model representation
Web Service Client Implementation

Textfeld: 1 Introduction
1.1 Motivation
The approach of integrating structures for Content Based Image Retrieval (CBIR) in the multimedia middleware framework METIS is prominent due to the fact that providing this feature is essential for certain audio-visual application scenarios for which the system serves as basis. An introduction to this topic and brief knowledge is already available through former plugin development projects, enabling the system to track duplicate images in its repository. However as the
implemented mechanism can neither be easily extended to provide a broader range of feature characterization, nor is it based on a common data model, there is a need to bring the CBIR approach to the next level. That means to offer support for emerging standards like MPEG-7, the Multimedia Content Description Interface and its rich set of standardized description tools as well as provide mechanisms for automatically or semi-automatically extracting them from the system’s given AV material. This data can then be used for building
description schemes for multimedia classification or for creating feature based queries.

In the stage of doing research on the field of CBIR tools and extraction algorithms, the eXperimentation Model, a non-normative reference software implementation for MPEG-7 was identified to mostly match our requirements as it offers applications for triggering MPEG-7 compliant feature extractions and contains a good set of implemented algorithms. Although the software was used for proof of concept in the standardization process it lacks of detailed documentation on its configuration - even getting it to work is a challenge. However, properly configuring the settings is mandatory for receiving suitable results.
Therefore all available information on the framework including hints from expert correspondence are aggregated in this paper to give a complete overview on the XM for setting up its infrastructure and for understanding its design.
The architecture of the XM framework does not allow directly adding an interface for automatically using its applications, so a solution for wrapping the service is required. Due to various limitations the eXperimentation Model has from its underlying hardware and software, as well as the fact that there is actually no need to run the METIS instance on the same machine the XM is available on, a solution for separating the components is presented. Therefore the main part of the software implementation process concerns the creation of a multi-client server architecture for offering the MPEG-7 feature extraction functionality with a well-defined interface and a platform-independent technology.

Finally in addition to the METIS aspect, the "feature-extraction as web-
service" also very well fits the BRICKS approach, a distributed, service oriented architecture for European cultural institutions, where the work is expected to be continued. The resulting web service solution, however, is not restricted to BRICKS but has the main advantage of being accessible by any application which wants to make use of the XM for MPEG-7 feature extraction functionality.
Textfeld: 1.2 Thesis Outline
The requirements for this master thesis are to offer support for a basic set of MPEG-7 Visual descriptors by extending METIS through its plugin mechanism.
For extracting feature characteristics from its attached media the MPEG-7 reference software is used, which is made available as a stand-alone component by using the web service technology, including functionality for remote data handling and a wrapper to set up its external infrastructure. The created semantic pack must be capable of capturing the description results. On basis of this
provided data further CBIR functionality can be built.

First of all, it is important to have an insight into the main technologies upon which the practical work is built. METIS is the core technology the developed solutions extend and are dependent on - it is described first. The paper provides a brief introduction to its main concepts as e.g. semantic packs and plugins for extending the core functionality and tries to depict the system’s design approach of being flexible at every level.
Afterwards we point out why there is an actual need for products like MPEG-7 and subsequently we present the Multimedia Content Description Interface. Its normative elements as well all Visual Descriptors for capturing multimedia characteristics at various abstraction levels are described in detail. Finally a short comparison to other textual description standards is given.
The chapter about the MPEG-7 reference software XM concentrates on assembling all rarely available information about obtaining, compiling and configuring the framework, containing valuable hints we retrieved through communication with several expert groups. Finally a qualitative analysis is presented
on how efficiently the MPEG-7 Descriptors cover their intended application domains in the task of distinguishing different media content. The last part of this section presents a selection on current projects and interesting tools built on either XM or the Multimedia Content Description Interface. As currently no overal picture on working groups developing with the standard is available the
presented resources may be helpful for carrying out further research activity.
Finally a brief introduction to the web service technology completes the overview of underlying technologies. It describes the SOAP Engine Apache AXIS for integrating web service support into Java and presents reasonable extensions for transferring binary data via SOAP.

The second part of this paper then deals with the designed software solutions, i.e. the XM Server - a web service implementation of the eXperimentation Model including support for multi-client and remote file handling, the XMFE METIS
plugin - a client software for integrating feature extraction support for METIS, and the XMBasic semantic pack - a data model which is capable of capturing the extracted data. The software is described by outlining its main design paradigms, its implemented functionality as well as the problems it is currently su¤ering from. For a better understanding, UML diagrams are used.
Textfeld: Please feel free to contact me when you‘re either interested in deploying or extending the created Java XM Wrapper Software for making the eXperimentation Model (XM) feature extraction functionality available as a web.
For further details and legal issues please see the written master thesis.