Image steganalysis and steganography in the spatial domain

In this dissertation, we propose different novel techniques both to detect hidden information (steganalysis) and to hide information (steganography). These techniques are presented in the form of a collection of five contributions, but sharing a common research problem. The first contribution presen...

Descripción completa

Detalles Bibliográficos
Autor: Lerch-Hostalot, Daniel
Tipo de recurso: tesis doctoral
Estado:Versión publicada
Fecha de publicación:2017
País:España
Institución:CBUC, CESCA
Repositorio:TDR. Tesis Doctorales en Red
OAI Identifier:oai:www.tdx.cat:10803/687395
Acceso en línea:http://hdl.handle.net/10803/687395
Access Level:acceso abierto
Palabra clave:esteganografia
esteganografía
steganography
estegoanàlisi
estegoanálisis
steganalysis
aprenentatge automàtic
aprendizaje automático
machine learning
processament d'imatges
procesamiento de imágenes
image processing
privadesa
privacidad
privacy
Tecnologies de la informació i de xarxes
004
Descripción
Sumario:In this dissertation, we propose different novel techniques both to detect hidden information (steganalysis) and to hide information (steganography). These techniques are presented in the form of a collection of five contributions, but sharing a common research problem. The first contribution presents three different methods to detect histogram shifting data hiding techniques, some of which are targeted attacks to specific schemes, whereas others are more general. As a second contribution, in the area of machine learning steganalysis, we present a novel feature extractor to detect information hidden in the spatial domain, which can be used as an additional submodel in the rich models framework, and which outperforms the accuracy of the state-of-the-art steganalysis by subtractive pixel adjacency matrix (SPAM) with fewer features. In the same context, the third contribution is a steganographic algorithm that exploits the weakness of some submodels to deal with high dimensional data (which typically use a threshold to overcome the dimensionality problem). As a fourth contribution, we present a new framework for unsupervised steganalysis with accuracy higher than the supervised methods in the state of the art, while bypassing the cover source mismatch (CSM) problem. Finally, as a fifth contribution, we present a novel approach to address the CSM problem based on the set of machine learning techniques known as manifold alignment.