Region Based Segmentation of Quickbird Imagery through Fuzzy Integration

The automatic segmentation of land cover features, within very high resolution (VHR) satellite imagery, is a complex task which is important to geo-spatial applications such as urban planning, crop monitoring and change detection. The dynamic grey-value variety of VHR imagery, along with environmental interference factors, such as cloud cover and poor lighting, impede the automation of land cover segmentation. The Fuzzy Band Ratio Hierarchical Split Merge Refinement (FBR HSMR) algorithm (Wuest and Zhang, 2008) presents a successful method for land cover segmentation through well known Band Ratios and Fuzzy Logic based comparison measures using the region-based Hierarchical Split Merge Refinement (HSMR) algorithmic framework. This paper is the presentation of an attempt to improve the automation of the FBR HSMR. In this approach, class development for region description and comparison is dynamically determined in contrast to static class development through Band Ratios. Fuzzy Adaptive Resonance Theory (ART) is employed for dynamic class development because of its unsupervised self-organizing capabilities and ability to estimate classes without initial estimates. In addition, users can control input to class development through input vector type selection. It is hypothesized that this approach will i) improve the automation of the FBR HSMR segmentation methodology and ii) expand the capabilities of the FBR HSMR to provide land cover segmentation to a wider range of satellite image scenes. * Corresponding author.