Quality measures for interactive image retrieval with a performance evaluation of two 3×3 texel-based methods

  • D. P. Huijsmans
  • M. S. Lew
  • D. Denteneer
Session 9: Image Databases
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1311)


The aim of the Leiden Imaging and Multi-media Group in collaboration with Philips is to develop and evaluate content-based indexing and interactive retrieval methods for large photo collections and to integrate them with annotation based methods. Ground-truth is provided by copy pairs in the Leiden Portrait Database, a database of scanned-in images of 19th-century Dutch studio portraits (“Cartes de Visite” ).

Our highly effective projection vector indexing method is compared with Virage Datablade and two binary texel (3x3 B/W patterns) statistic feature vectors: a reported well-performing Local Binary Pattern and our 2D binary gradient pixel Trigram.

Evaluation criteria, based upon the number of copies found back within the visible top [2logn] ranks, were defined for interactive internet image retrieval and applied to the ranking results for the test-set of 50 copy and 12 similar pairs embedded in 5570 portraits and studio logo images. Our evaluation shows that the projection method beats the binary texel based methods and Virage Datablade; the Trigram method performs better than the LBP method. Feature vector length reduction by grouping texel patterns in symmetry groups reduces the strenght of the Trigram method, whereas a KLT transform can be used to reduce the length of each feature vector by an order of magnitude without affecting the performance.

At “http://indl56b.wi.leidenuniv.nl:2000/" our visual search demo can be tried.


  1. [Huijsmans96a]
    Huijsmans D.P., Lew M.S.; Efficient content-based image retrieval in digital picture collections using projections: (near)copy location; in Proc. 13th ICPR Vienna 1996 vol III pp 104–108.Google Scholar
  2. [Huijsmans96b]
    Huijsmans D.P., Poles S., Lew M.S.; 2D pixel trigrams for contentbased image retrieval; in (eds)Smeulders A.W.W, Jain R.; Image databases and Multi-Media search; Proc.lth Int workshop IDB-MMS A'dam 1996 pp 139–145.Google Scholar
  3. [Lew96]
    Lew M.S., Huijsmans D.P., Denteneer D.; Content-based image retrieval: KLT, projections or templates; in (eds)Smeulders A.W.W, Jain R.; Image databases and Multi-Media search; Proc.lth Int workshop IDB-MMS A'dam 1996 pp 27–34.Google Scholar
  4. [Ojala96]
    Ojala T., Pietikainen M and Harwood D.; A comparative study of texture measures with classification based on feature distributions; Pattern Recognition vol 29-1 Jan 1996 pp 51–60.CrossRefGoogle Scholar
  5. [Wang90]
    Wang L., He D.C.; Texture classification using texture spectrum; Pattern Recognition vol 23 (1990) pp 905–910.CrossRefGoogle Scholar
  6. [Wood96]
    Wood J.,;Invariant pattern recognition: a review; Pattern Recognition vol 29-1 Jan 1996 pp 1–18.CrossRefGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1997

Authors and Affiliations

  • D. P. Huijsmans
    • 1
  • M. S. Lew
    • 1
  • D. Denteneer
    • 2
  1. 1.Computer Science DepartmentUniversity of LeidenRA Leiden
  2. 2.Philips Research LabEindhovenThe Netherlands

Personalised recommendations