Exploring the Color Histogram's Dataspace for Content Based Image Retrieval

• Author(s) / Creator(s)

  • Coman, Alex

• Technical report TR03-01. With the growth of available information in digital format, indexing has drawn much attention as a viable solution to reduce retrieval time when searching large databases. There are many indexing techniques available nowadays, but, as they were developed with general goals in mind, they do not perform at their best in many cases. In this report we explore the use of prior knowledge about the data to be indexed (e.g. its spatial distribution) in order to enhance the performance of the indexing structure. Based on a concrete example -- the constraint existing in color histograms -- we explore how such property induces other data constraints. We design a specialized index structure to take advantage of the induced data constraints. We also propose a new lower bound distance between data points and Minimum Bounding Rectangles to be used when indexing histograms with existing indexing structures. The proposed indexing technique is shown to be efficient and stable with respect to variations in number of data objects, page size, feature vector dimensionality and variation in number of nearest neighbors to be searched for. | TRID-ID TR03-01

• Date created

  2003

• Subjects / Keywords

  • Content based image retrieval
  • Indexing techniques
  • Color hisotgrams

• Type of Item

  Report

• DOI

  https://doi.org/10.7939/R3416T300

• License

  Attribution 3.0 International