Underwater Images of Stony Reefs in Turbid Temperate Waters: A Dataset for Evaluating Textural Descriptors

The data presented here were collected from 33 sampling sites in the coastal and offshore reefs of the southeastern Baltic Sea (depths ranging from 4.4 to 42.2 m) during the summer-autumn sampling seasons of 2022–2023. This dataset includes 124 video frames (divided into training and testing), 2,800 patches (100x100 px), with 350 patches for each class, along with annotated data. Five key biological classes representing different textures were selected: red algae Furcellaria lumbricalis, brown algae Vertebrata fucoides, green algae Cladophora sp., and large and small blue mussels (M. edulis trossulus). Additionally, three lithological classes were chosen: stable substrates consisting of cobbles (6–25 cm) and boulders (>25 cm), pebbles (0.2–6 cm), and sand (<0.2 cm). The testing dataset was used to create annotated data, which include pixel-wise segmentation of 165 images, annotated by three benthic experts. The extracted patches from the frames were used for image classification with CatBoost based on textural descriptors, while the annotations served to validate the classification results. All scripts used for patch and image classification tasks are provided here. Additionally, 16 algorithms from the Turan (2018) collection were reimplemented in C++ at CCOM/JHC and organized as Python modules for ease of use. All statistical analyses related to this study can be found in the R script Statistical_analysis.R.

Turan, C., Lam, K.-M. 2018. Histogram-based Local Descriptors for Facial Expression Recognition (FER): A comprehensive Study. J. Vis. Commun. Image R.. https://doi.org/10.1016/j.jvcir.2018.05.024.