Signature Quadratic Form Distance

Basic data for this project

Type of project: Own resources project
Duration: 01/04/2009 - 30/09/2013

Description

The Quadratic Form Distance has been proposed in the 1990s as a method of modeling color-based image similarity. This distance has become prominent since its utilization within IBM’s QBIC project, where it has been investigated for efficient and effective content-based retrieval in large image databases. Initially defined as a distance between color histograms, the Quadratic Form Distance has been researched intensively and adated to many different application domains. In this research project, we develop and investigate the Quadratic Form Distance for adaptive-binning feature histograms, which are also known as feature signatures. The resulting Signature Quadratic Form Distance is an adaptive generalization that can be utilized as a flexible distance-based similarity model in any application domain. In this project, we aim to expand the applicability of the Signature Quadratic Form Distance to more complex application settings. For this purpose, we research various methods to increase the processing performance of this distance in single-core CPU and multi-core CPU environments as well as on many-core GPU architectures. In addition, we have also shown how to utilize the Signature Quadratic Form Distance for continuous feature distributions as a measure of distance between Gaussian mixture models of high-dimensional feature descriptors.

Keywords: Signature Quadratic Form Distance