Open framework for the analysis of rich media

Martins, Filipe (2011) Open framework for the analysis of rich media. (MSc(R) thesis), Kingston University, .

Abstract

This thesis describes the work undertaken to complete a two year programme of study to investigate an 'open framework for the analysis of rich media' (OpenFARM), under the Knowledge Transfer Partnership (KTP) scheme. The broadcast industry has many applications for media analysis, and in particular, for automated indexing, to facilitate the searching and browsing of content. One challenge is the design and implementation of a framework that enables any given broadcast system to integrate any chosen analysis component. Such a framework would benefit the broadcast user, allowing them to add to their tool set without significant disruption, and it would also provide more opportunities for any given tool to be deployed. The key requirements are: a simple execution environment for analysis components, in a robust, scalable system which does not proliferate the number of databases on which the Broadcast System depends. This thesis contains eight chapters that provide an analysis of the problem to be solved, a review of previous work, the description of an appropriate design to solve the problem, and the development, test and use of this solution. The assessment of the problem demonstrates the issues at stake, principally the current mismatch between broadcast and analytics operating environments. The literature review surveys the previous work in three relevant areas: software engineering, computer vision and state of the art comparable systems. A fundamental requirement is the capability to be integrated with any computer vision analysis tool and any Media Asset Management (MAM) system regardless of their platform and native programming language. Chapter IV reviews how this type of interoperability is achieved and the ontology and concepts adopted. The proposed solution comprises three stages. The first, termed "Analysis Stage" which is deemed to deal with low level signal processing analysis tools, the second characterized as "Management Stage" which acts as a bridge between the framework and the third stage, the "Deployment Stage" that relates to all the possible third-party connecting applications. All of these stages are described in depth in Chapter V in which a description about the framework payload, the concept of generic pojo and the distributed architecture the framework comprises is explained. In order to evaluate the framework and test it end-to-end, a few analysis tools such as a scene cut detection, flash photography, fade to black detector and text recognition plug-ins were developed. Such plug-ins were also not picked at random and Chapter VI shows how important they are in the industry. Finally the thesis concludes with an overall evaluation of the framework and a revision for future work.

Actions (Repository Editors)

Item Control Page Item Control Page