DSpace UNL

RUN >
Faculdade de Ciências e Tecnologia (FCT) >
FCT Departamentos >
FCT: Departamento de Informática >
FCT: DI - Dissertações de Mestrado >

Please use this identifier to cite or link to this item: http://hdl.handle.net/10362/5963

Title: Automated image tagging through tag propagation
Authors: Silva, Miguel Marinhas da
Advisor: Magalhães, João
Keywords: Multimedia annotation
Machine learning
Multimodal classification
Tags
Metadata extraction
ImageCLEF
Issue Date: 2011
Publisher: Faculdade de Ciências e Tecnologia
Abstract: Today, more and more data is becoming available on the Web. In particular, we have recently witnessed an exponential increase of multimedia content within various content sharing websites. While this content is widely available, great challenges have arisen to effectively search and browse such vast amount of content. A solution to this problem is to annotate information, a task that without computer aid requires a large-scale human effort. The goal of this thesis is to automate the task of annotating multimedia information with machine learning algorithms. We propose the development of a machine learning framework capable of doing automated image annotation in large-scale consumer photos. To this extent a study on state of art algorithms was conducted, which concluded with a baseline implementation of a k-nearest neighbor algorithm. This baseline was used to implement a more advanced algorithm capable of annotating images in the situations with limited training images and a large set of test images – thus, a semi-supervised approach. Further studies were conducted on the feature spaces used to describe images towards a successful integration in the developed framework. We first analyzed the semantic gap between the visual feature spaces and concepts present in an image, and how to avoid or mitigate this gap. Moreover, we examined how users perceive images by performing a statistical analysis of the image tags inserted by users. A linguistic and statistical expansion of image tags was also implemented. The developed framework withstands uneven data distributions that occur in consumer datasets, and scales accordingly, requiring few previously annotated data. The principal mechanism that allows easier scaling is the propagation of information between the annotated data and un-annotated data.
Description: Trabalho apresentado no âmbito do Mestrado em Engenharia Informática, como requisito parcial Para obtenção do grau de Mestre em Engenharia Informática
URI: http://hdl.handle.net/10362/5963
Appears in Collections:FCT: DI - Dissertações de Mestrado

Files in This Item:

File Description SizeFormat
Silva_2011.pdf1.99 MBAdobe PDFView/Open
Statistics
View Statistics
FacebookTwitterDeliciousLinkedInDiggGoogle BookmarksMySpaceOrkut
Formato BibTex mendeley Endnote Logotipo do DeGóis 

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

 

Universidade Nova de Lisboa  - Statistics  - Feedback
Estamos no RCAAP Governo Português separator Ministério da Educação e Ciência   Fundação para a Ciência e a Tecnologia

Financiado por:

POS_C UE