Sketch-based image retrieval using keyshapes

Jose M. Saavedra*, Benjamin Bustos

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

37 Scopus citations


Although sketch based image retrieval (SBIR) is still a young research area, there are many applications capable of exploiting this retrieval paradigm, such as web searching and pattern detection. Moreover, nowadays drawing a simple sketch query turns very simple since touch screen based technology is being expanded. In this work, we propose a novel local approach for SBIR based on detecting simple shapes which are named keyshapes. Our method works as a local strategy, but instead of detecting keypoints, it detects keyshapes over which local descriptors are computed. Our proposal based on keyshapes allow us to represent the structure of the objects in an image which could be used to increase the effectiveness in the retrieval task. Indeed, our results show an improvement in the retrieval effectiveness with respect to the state of the art. Furthermore, we demonstrate that combining our keyshape approach with a Bag of Feature approach allows us to achieve significant improvement with respect to the effectiveness of the retrieval task.

Original languageEnglish
Pages (from-to)2033-2062
Number of pages30
JournalMultimedia Tools and Applications
Issue number3
StatePublished - 29 Oct 2014
Externally publishedYes

Bibliographical note

Publisher Copyright:
© 2013, Springer Science+Business Media New York.


  • Content-based image retrieval
  • Local descriptors
  • Local matching
  • Sketch-based image retrieval


Dive into the research topics of 'Sketch-based image retrieval using keyshapes'. Together they form a unique fingerprint.

Cite this