Performance Comparison of Flood Image Segmentation Techniques

Effective segmentation of flood-affected areas must be done for effective disaster management and recovery planning. This paper conducts a performance comparison of four different segmentation techniques on flood-area images. Techniques adopted in this work are region growing, active contour, histogram, and K-means clustering. K-means algorithm is based on Machine Learning technique. Further, consistency verification is applied on segmented images. The original flood image is segmented first using four methods. Generated masks are compared with a ground truth mask. The Score value is computed using a confusion matrix. Difference maps are further generated to visualize segmentation results. It provides valuable insights into the strengths and weaknesses of each method that can be used to select optimal approaches for the analysis of flood areas. Score is considered the performance metric used to assess the performance of various techniques. Machine learning based K-means segmentation along with consistency verification outperforming than other techniques.