“…A set of low‐level visual descriptors is selected from the descriptor bank of ParaDISE (Schaer, Markonis, & Müller, ; Markonis et al, 2017) and their combination is explored (García Seco de Herrera, Markonis, Schaer, et al, 2013) to optimize the outcomes. The following descriptors are chosen after the performance tests on a different ImageCLEF database; - Bag of Visual Words (BoVW) using the Scale Invariant Feature Transform (SIFT) ( Lowe, ) with a spatial pyramid matching ( Lazebnik, Schmid, & Ponce, ) (BoVW‐SPM) —each image is represented by a histogram symbolizing a set of local descriptors represented in visual words from a previously learned vocabulary; spatial information is added to the BoVW‐SIFT descriptor;
- Bag of Colors (BoC) ( García Seco de Herrera, Markonis, & Müller, ) with an n × n spatial grid (Grid BoC) —each image is represented by a histogram symbolizing the colors from a previously learned vocabulary; spatial information is added to the BoC descriptor;
- Color and Edge Directivity Descriptor (CEDD) (Chatzichristofis & Boutalis, )—color and texture information is produced by a 144 bin histogram. Little computation is needed for its extraction;
- Tamura texture (Tamura, Mori, & Yamawaki, )—this descriptor extracts six visual properties: coarseness, contrast, directionality, line‐likeness, regularity, and roughness.
…”