Choosing geometric dissimilarity measure for content based coral reefs image retrieval

Abstract

The dissimilarity measure used by a Content-Based Image Retrieval (CBIR) system significantly affects its performance. Choosing the right dissimilarity measure is important especially when we have large low-level features to represent each image in the database. This paper presents the performance of various geometric distance measures for retrieval of images from a coral reefs database that consists of three groups of coral. Based on the results obtained by Precision-Recall graphs, there is no single distance measure that best for all queries. Therefore, Mean Average Precision is used to measure the overall performance, and the results showed that the top three best geometric distance measures for retrieving images from a coral database are the Squared Chord, City Block, and Canberra.