SAPIR (Search In Audio Visual Content Using Peer-to-peer IR)

SAPIR is an EU FP6 project co-funded by the "Networked Media" Unit and is conducted by 9 partners. It aims at extending the power of web searches beyond centralized text and metadata to include distributed audio-visual content. Implementing real content-based, audio-visual searches requires mediaspecific understanding and extremely high CPU utilization, which do not scale in today's centralized solutions. SAPIR aims at breaking this technological barrier by developing a large-scale, distributed peer-to-peer architecture that makes it possible to search audio-visual content using the query-by-example paradigm. To achieve this goal, a common framework for feature extraction from all media contents is being developed allowing similarity search and ranking along all supported media.

Collaboration:

The collaboration with SAPIR mainly concerns the exploitation of the D4Science e- Infrastructure.

Service

In order to measure and validate the quality of the developed search engine, SAPIR has initially planned an experimentation with 100 Millions images gathered from Flikr. Extracting features from this large collection is very computational intensive, therefore the processing capacity needed to extract features from this large collection was exceeding the resources available to the project partners. In order to overcome this problem, SAPIR established a collaboration with DILIGENT and with its successor, D4Science, for the hosting of the feature extraction services on the existing infrastructure. By exploiting the D4Science infrastructure, 22 Millions of images were processed in the period August–September 2008. This processing has generated 67 Millions of objects that use approximately 2.7 TB of storage.

Project web site:

http://www.sapir.eu/

»»