Image Segmentation Techniques for Urban Land Cover Segmentation of VHR Imagery: Recent Developments and Future Prospects

Abstract

Since the last decade, remote sensing (RS) sensor technology has seen a steady development, which has refined spatial detail of satellite imagery close to sub-meter resolution of aerial imagery. The development also brought a paradigm shift, from pixel-based to object-based image analysis (OBIA), in remote sensing image processing. OBIA tries to mimic human perception of images as geographic objects. The basic step of geographic object formation in OBIA is image segmentation. Fortunately, image segmentation has been researched in computer vision for the last four decade. However, this doesn’t alleviate the segmentation problem because image segmentation is domain specific. This paper reviews image segmentation techniques in the domain of urban land cover segmentation of very high spatial resolution (VHR) satellite imagery. The paper categorizes the segmentation techniques into eight categories namely, clustering, level-set, Markov random field, fuzzy logic, neural network, multi-scale, watershed, and hierarchical split and merges (HSMR). The paper also describes the recently developed techniques, deduces trends, (e.g., widely used techniques and commercially developed techniques) and elaborates on the potential techniques, where a researcher can dig in. The paper concludes that multi-scale and watershed based techniques are the most appropriate for OBIA of VHR images of urban areas.