Modified Fast and Robust Fuzzy C-means Algorithm for Flood Damage Assessment using Optimal Image Segmentation Cluster Number

The issue of flooding has been one of the common global natural disasters that leads to costly property destruction and hazardous situations. Fortunately, unmanned aerial vehicle imagery advancements have paved the way for assessing flood damages. Raw images taken may further be processed through image segmentation that is done by applying the Fuzzy C-means method. This paper modifies the fast and robust Fuzzy C-means algorithm to account for the issues regarding the need for stronger assumptions on the initial cluster number. Three validity coefficients, namely the Partition Coefficient, Xie and Beni's Index, and the Silhouette Coefficient, were calculated to obtain ideal image partitioning. Applying the proposed algorithm to the aerial images from the AIDER dataset, results have shown how the optimal cluster number varied depending on the elements of the image being segmented. Nonetheless, the algorithm addresses the need for having strong initial assumptions while being able to minimize computational complexity.