| |
Pattern Recognition and Semantic Image Description
Goal: This research topic aims to examine new approaches to pattern recognition and semantic image description
The multimedia content description tools standardized by MPEG-7 are thereby used as basement for new algorithms. The MPEG-7 features are used in selected image identification scenarios to detect specific image contents. Object-based MPEG-7 metadata are also extracted to achieve semantic scene descriptions for given application-specific classes of video sequences.
Structure of the Semantic
Image Retrieval Approach
High-level retrieval systems for semantic image retrieval are typically very complex. They describe, archive, and if needed provide the user with multimedia data based on semantic retrieval criteria. The principle of such a search engine is depicted in Fig. 1.
Fig. 1: Principle of a semantic content-based image search engine
Image and video content is described using a Semantic Texture Classificator (STC). The metadata extracted by the STC are stored in a database. The Semantic Texture Descriptor (STD) links the database to the search engine developed at FhI-HHI. STD functionalities enable user interaction to enhance the quality of search results. STD uses the metadata extracted by STC and analyzes local and global image features.
Search Engine with Integrated Semantic
Features
A screen shot of the FhI-HHI search engine is depicted in Fig. 2. The search engine features the following functionalities:
- Navigation through image databases or search results,
- Feature (e.g. STD) extraction from a single or a list of images
- Similarity- and content-based image query.
Feature extraction can be selected via the „Extract“ button, while navigation can be enabled through the “Browse” option and can be carried out randomly or sequentially.
Fig. 2: User interface of the FhI-HHI search engine
Image query can be done based on different criteria (color, texture, STD etc.) that can be selected via the “Search” option. The large image on the left hand side (cf. Fig. 2) represents the query image selected by the user. The latter can also select specific semantic texture combinations he wishes to be found. These are highlighted by red dots in the query image (cf. Fig. 2). The small images on the right hand side represent the search result ordered in decreasing similarity to the query image.
Publications
Publications in Journals
2002
- Karsten Müller and Aljoscha Smolic:
MPEG-7: Anwendungen für die Film- und Fernsehtechnik - das Multimedia Content Description Interface,
FKT- Fachzeitschrift für Fernsehen, Film und Elektronische Medien, pp. 703-705, December 2002, in German.
- Karsten Müller:
MPEG-7: Content Description of Multimedia Data,
Production Reality - Journal for Broadcast, Post-Production, Animation and Content Description, November 2002.
Publications in Conference Proceedings
2004
-
Patrick Ndjiki-Nya, Oleg Novychny, and Thomas Wiegand:
Video Content Analysis Using MPEG-7 Descriptors,
1st European Conference on Visual Media Production (CVMP), London, United Kingdom, pp. 95-101, March 2004.
(Download as
pdf
)
|
Start
Organisation