University of Twente Student Theses

Login

Automatic classification of television commercials

Nijmeijer, Tom (2008) Automatic classification of television commercials.

[img] PDF
2MB
Abstract:The last years we have seen a rapid growth of the World Wide Web and vastly increasing amounts of multimedia on it. As the volume of multimedia increases it becomes increasingly important to be able to efficiently search through this multimedia. Unfortunately searching through multimedia is less straightforward than searching through text data. Most commonly search queries are expressed in words, which match text documents in format. Multimedia documents on the other hand are not inherently expressed in words. Thus techniques have to be found to that can match textual search queries to the data available in multimedia documents. In this thesis we explore the possibility of using image and video analysis techniques to classify television commercials. If we can generate proper classifations for the television commercials we can then search through them using normal textual search queries. Classifying television commercials is a complex process. In a number of consecutive steps the raw video data has to be filtered down in order to eventually allow a machine learning algorithm to learn the proper classiffcations. Using the available literature an approach for this filtering was devised. The approach starts with the detection of shot boundaries, followed by the extraction of a keyframe from each detected shot. After that keypoints are selected in each keyframe and described by SIFT descriptors. A clustermodel is trained on the entire collection of SIFT descriptors to obtain a vocabulary of 'visual' words. A histogram of the 'visual' words is then constructed for each keyframe. These keyframe histogram representation are used by a machine learning algorithm to train a classifier.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Interaction Technology MSc (60030)
Link to this item:http://purl.utwente.nl/essays/57880
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page