Signature Quadratic Form Distance for Efficient Multimedia Database Retrieval

• Applicant: Professor Dr. Thomas Seidl
• Subject Area: Security and Dependability, Operating-, Communication- and Distributed Systems
• Term: from 2012 to 2016
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 211499611

The Signature Quadratic Form Distance is an adaptive distance-based similarity measure for the comparison of exible feature representations of multimedia data. This similarity measure is a generalization of the Quadratic Form Distance and enables the computation of similarity values among feature representations of different structure and size. In this way, it is applicable to complex multimedia data and furthermore comparable with the quality of other adaptive distance-based similarity measures. Within the scope of this project, we intend to investigate and increase the applicability and efficiency of the Signature Quadratic Form Distance in the field of content-based multimedia retrieval. We plan to develop a novel model of the Signature Quadratic Form Distance which is able to compare discrete as well as continuous feature distributions of multimedia data. Based on this novel distance model, we plan to investigate and extend the applicability of the Signature Quadratic Form Distance regarding bag-of-visual- word approaches in order to increase the latter's exibility. In order to process similarity queries efficiently, we finally plan to approximate discrete and continuous feature distributions as well as distance computations. Findings of the aforementioned topics will expand the applicability and improve the efficiency of the Signature Quadratic Form Distance to be leveraged in more complex settings regarding content-based multimedia retrieval.

• DFG Programme: Research Grants