Abstract
In this work, we present the novel Inter-GIP Distances (IGD) feature andits integration into the Gestalt Interest Points (GIP) image descriptor.With the ongoing growth of visual data, efficient image descriptor methodsare becoming more and more important. Several local point-based descriptionmethods have been defined in the past decades. Accuracy and descriptor sizeare important factors when selecting the appropriate method for a givenretrieval problem. The method presented in this work describes images withonly a few very compact descriptors. To test our descriptor, we developedan image classification prototype and conducted several experiments with apublicly available horses dataset and a food dataset. Our experiments showthat only a few of the very compact GIP image descriptors are necessary toquickly classify the images from the datasets with high accuracy.Furthermore, we compared our experimental results to state-of-the-art localpoint-based description methods and found that our method is highlycompetitive.
Reference
Hörhan, M., & Eidenberger, H. (2017). The Gestalt Interest Points Distance Feature for Compact and Accurate Image Description. In Proceedings IEEE International Symposium on Signal Processing and Information Technology (p. 5). http://hdl.handle.net/20.500.12708/57261