Introducing a new image dissimilarity measure with an application to character image clustering in degraded historical documents

  • Authors:

  • Sebastian Colutto

  • Affiliations:

  • University of Innsbruck, Innsbruck, Austria

  • Venue:
  • DAS '10 Proceedings of the 9th IAPR International Workshop on Document Analysis Systems
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper we present a novel method for the calculation of the distance between two input images that are representing characters of an historical document. The ultimate goal is to create a high quality clustering of the images, i.e. to extract an inventory of the document. Our image dissimilarity measure is based upon the Local Distance Map and robust curvature estimation using Integral Invariants. We demonstrate the superior behaviour of the image dissimilarity measure with experiments on three datasets of different font and quality comparing them to standard shape descriptors as well as clustering results produced by a state-of-the-art OCR engine.