Hierarchical Clustering of Natural Disaster Images Using Visual Features

This paper presents a hierarchical clustering method for natural disaster images. By displaying natural disaster images with similar categories, users are able to intuitively understand their surrounding environment and potential risks such as flood and sediment damage. Because there are unstructured natural disaster images without textual descriptions, a method that does not use textual features is necessary. The contribution of this study is to newly develop a method to hierarchically cluster natural disaster images using only visual features. Experimental results for images that are available from the research database show the superiority of our hierarchical clustering compared with flat clustering.