Powered by TikiWiki   Tue 25 of Apr, 2017 [06:45 UTC]
  • People


MUFIN Image Search

MUFIN image search screenshot The MUFIN Image Search system demonstrates the power of general search engine MUFIN (Multi-Feature Indexing Network). The mission of the MUFIN project is to develop a versatile indexing and searching system focusing on general similarity search in very large databases. All individual technologies incorporated in MUFIN are based purely on the metric model of data. Read more about MUFIN, its architecture, and its features.

Image dataset and similarity

The MUFIN image search system is an "instance" of MUFIN for content-based similarity search on large datasets of digital images. The image database has the following characteristics:

  • 50 million image database taken from CoPhIRexternal link
  • original images are from Flickrexternal link
  • similarity based on a combination of five MPEG-7 global descriptors:
    • ColorStructure (weight 3)
    • ColorLayout (weight 2)
    • ScalableColor (weight 2)
    • EdgeHistogram (weight 4)
    • HomogeneousTexture (weight 0.5)

Please, note that global descriptors, in general, capture the overall characteristics of the images. Adopting this approach, our similarity measure does not locate individual objects from the query image in other images but compares the images as a whole.

The weights of particular descriptors were determined experimentally trying to capture the human notion of similarity for non-specific images. As the system is based purely on the metric space model of data, we can use different weights and even different image charactersticss without touching the indexing and searching techniques.

Indexing and searching

The specific indexing and searching structure of this instance is formed by:

  • M-Chord - a distributed data structure which is based on the principles of peer-to-peer structures
    • 2000 logical peers
  • M-Tree data structure as local index structrure at each of the peers

See the Publications section for sources containing detailed information about these and other MUFIN technologies.

Hardware infrastructure

The hardware infrastructure has been provided by MetaCentrumexternal link and mainly by IBM Shared University Research Awardexternal link.

  • 32 CPU cores
  • 64GB RAM
  • 24 hard disks

See the Implementation section for implementation details.

Supported by

Faculty of Informatics
Faculty of Informatics,
Masaryk University, Brno


Czech Science Foundation

Search In Audio Visual Content Using Peer-to-peer IR
IST FP6 Project

National Research Program

IST FP6 Project