Generalizing the Simple Linear Iterative Clustering (SLIC) superpixels

Superpixels are a promising group of techniques allowing for generalization of spatial information. Among this group, the Simple Linear Iterative Clustering (SLIC) superpixels algorithm proved to be first-rate, both in terms of the quality of the output and the performance. SLIC, however, is limited to detecting homogeneous areas and uses the Euclidean distance only. Here, we propose an extension of SLIC allowing to use any specified distance measure for single or multi-layered spatial raster data. To present our idea, we use the extension to create an over-segmentation of areas with similar proportions of different land cover categories in Ohio. Given a proper distance measure, the proposed extension can also be used for other scenarios, including creating regions of similar temporal patterns or similarly ranked areas. Depending on the use case, the resulting superpixels could be either the result of the analysis or the input for further classification or clustering.