Presenting an Object-Based Approach Using Image Edges to Detect Building Boundaries from High Spatial Resolution Images

Although various building detection algorithms from high spatial resolution satellite (HSRS) images have been presented in recent years, there are yet some difficulties to detect building boundaries for mapping purposes. The present study aims to propose a new approach to detect building boundaries from HSRS images by focusing on higher detection rates. The approach utilizes the idea of object-based image processing. However, it has an innovative vision using image edges instead of traditional image segments as objects. To evaluate the efficiency of the proposed approach, two datasets which have different contrast between building and non-building areas, are used: the first dataset has a high contrast between building and non-building areas (HC) and second has a low contrast (LC). The results are compared with the results of two segmentation-based algorithms, i.e., classification based on edge-based segmentation (CBES) and classification based on multi-resolution segmentation (CBMS). The comparisons indicate higher efficiency of the proposed approach for the HC dataset with 6–14% higher detection rate (lower omission error) than the two segmentation-based algorithms. For the LC dataset, the proposed approach is already more efficient than CBMS with 10–25% higher detection rate. However, it has lower efficiency than CBES with 15–36% higher omission errors. Though, the proposed approach is generally more robust than CBES and CBMS algorithms based on standard deviation values of evaluation metrics.