Combined Features for Content Based Image Retrieval: A Comparative Study

Abstract

Multimedia resources are rapidly growing with a huge increase of visual contents. Thus, searching these images accurately and efficiently for all types of datasets becomes one of the most challenging tasks. Content-based image retrieval (CBIR) is the technique that retrieves images based on their visual contents. So that, selecting appropriate features that describe an image sufficiently is a clue for a successful retrieval system. To this end, in this paper, a comparative study to investigate the effect of using a single and a combined set of features in the context of a CBIR is presented. To achieve this goal, several features including, edge histogram (EHD), color layout (CLD) and fuzzy color texture histogram (FCTH) as well as different combinations of these features such as, all edges (local, global and semi-global edges), all edges with CLD and finally, all edges with FCTH have been exploited. To demonstrate the effectiveness of the proposed method, a set of experiments utilizing different images datasets have been carried out. The results in terms of precision, recall, F-measure and mean average precision show a higher retrieval accuracy while using a set of combined features compared to exploiting only single features for the same retrieval task.