Image Search using Dominant Colors

Article type: Research Article

Authors: Galiński, Grzegorz | Skarbek, Władysław

Affiliations: Multimedia Laboratory, Institute of Radioelectronics, Warsaw University of Technology, Nowowiejska 15/19, 00-665 Warsaw, Poland. E-mail: G.Galinski@ire.pw.edu.pl; W.Skarbek@ire.pw.edu.pl

Abstract: In this paper, the use of the Dominant Color descriptor, one of the visual descriptors defined in MPEG-7 standard for image indexing and search, is discussed. Various similarity functions for matching the descriptors are presented. Next, the concept of dominant colors cumulative distribution function (CDF) is introduced and a new distance function for the descriptors matching is also proposed. The function is equivalent to the computing volume of space bounded by two dominant colors CDFs. Efficient algorithm for computing CDF-based distance which uses the properties of discrete CDF function, is also presented. Finally, various dominant colors matching methods are compared in a similarity image search scenario consisting of 'query-by-example' nearest neighbours search, and the experimental results are presented and discussed.

Keywords: image search, particle ensembles, dominant colors, MPEG-7

Journal: Fundamenta Informaticae, vol. 82, no. 1-2, pp. 105-112, 2008