The Role of Quantitative Metrics in Enhancing Spatial Information Retrieval via Fuzzy K-Means Clustering

Remote sensing information is critical to preserve environments in tropical and mountain regions (e.g., costal hurricane mitigation or mountain forest management). In order to partition graphical information into meaningful regions and extract salient objects, segmentation is used to identify visual signatures based on similarity criteria. Due to atmospheric dispersing, nonlinear denoising is needed to capture intrinsic information. To identify illustrious objects and regions, image segmentation using K-Means clustering is generally applied to partition information into diverse clusters. Since each pixel might have certain degree of belongings to multiple clusters, fuzzy K-Means clustering is introduced to depict belongings using fuzzy membership functions, which classifies pixels into two or more clusters. Without general rules to determine an optimal number, the key problem is to specify the desired number of clusters. Quantitative measures are also proposed to identify the actual number of clusters to enhance decision support accuracy and optimize K-Means clustering.