Semi-Automatic Extraction of Large and Moderate Buildings from Very High-Resolution Satellite Imagery using Active Contour Model

Abstract

The satellite imagery has different types of objects like buildings, vegetation, lakes, roads, grounds etc. Each type of object has unique characteristicslike texture, colour,spatial arrangement etc that differentiates them from other types of objects. Using a single method to extract all these types of objects might not be logical. Detecting buildings from satellite imagery is of major importance for supporting government related activities and a great support in crisis situations for disaster management. This work presents a new way of automating the extraction of buildings from high resolution satellite images using Object Based Image Analysis (OBIA). The proposed algorithm utilizes an active contour model called Chan-Vese segmentation that is capable of accurately locating various objects in the image. Additionally, to ensure accuracy, the algorithm employs Normalized Difference Vegetation Index (NDVI) mask to remove vegetation areas. Moreover, these detected objects pass through a careful filtering process based on regional characteristics like minimum area and object width thereby improving the precision of the identified structures. The correctness of results is confirmed by conducting rigorous quantitative assessments for validation purposes. This means that not only does this comprehensive approach guarantee error-free building identification but also suggests a strong framework for automatic building extraction from high resolution satellite imagery.