Mediana: Methods and Tools for Scientific Media Analysis

Keywords: image and video analysis, content-based multimedia retrieval, computational media aesthetics, distributed multimedia processing, service-oriented Grid computing

The Mediana project is conducted within a large trans-disciplinary research center located at the University of Siegen, Germany, entitled "Media Upheavals". It examines the prerequisites and structures of the media upheavals at the beginning of the 20th century ("analogous media") and in the crossover to the 21st century ("digital media"), in terms of their meaning for the emergence and transformation of media cultures, and for the development of media aesthetics – while considering the technological factors involved. A guiding premise of the investigation is the observation that media upheavals suppose comprehensive, discontinuous, structural changes within media history.The expression "media upheaval" can include technological factors and anthropological dimensions of media history as well as inherited social and cultural institutions or aesthetic traditions.

The tools developed in the Mediana project to support this kind of research can be divided into two groups:

1. Internet-based software environments to enable the management of heterogeneous data sources and support the scientific working processes of media researchers,
2. Tools to search and retrieve particular objects and events in audiovisual data and WWW documents.

Our current work focuses on methods for recognizing relevant objects in images and videos, understanding events and their relationships in multimedia data, and building distributed software architectures based on the service-oriented Grid computing paradigm for multimedia processing and retrieval. In particular, we have developed image and video content analysis algorithms for

• shot boundary detection
• scene detection
• person indexing in video (face detection and recognition, speaker classification)
• camera motion estimation
• distinction of camera and object motion
• detection of moving objects
• text detection and recognition in images and video
• improving robustness of video content analysis algorithms with respect to
  • compression
  • parameter settings

Recently, we have started investigating approaches to bridge the "semantic gap" between such computable low-level features and the high-level semantics which are relevant for the user ("Computational Media Aesthetics"). The ultimate goal is to provide support for understanding the meaning of audiovisual objects and events.

Funding: Deutsche Forschungsgemeinschaft (DFG, PAK 509)

Contact: Ralph Ewerth, Markus Mühling