A Supervised Building Detection Based on Shadow using Segmentation and Texture in High-Resolution Images

Building detection in aerial or satellite imagery is one of the most challenging tasks due to the variety of shapes, sizes, colors, and textures of man-made objects.To this end, in this paper, we propose a novel approach to extracting buildings in high-resolution images based on prior knowledge of the shadow position.Firstly, the image is split into superpixel patches; the colors and texture features are extracted for those patches.Then using the machine learning method (SVM), four classes are made: buildings, roads, trees, and shadows.According to the prior knowledge of shadows position, a seed point initial has been defined along with an adaptive regional growth method to determine the approximate building location.Finally, applying a contouring process included an open morphological operation to extract the final shape of buildings.The performance is tested on aerial images from New Zealand area.The proposed approach demonstrated higher detection rate precision than other related works, exceeding 97% despite the complexity of scenes.

• building detection
• high-resolution aerial imagery
• superpixels
• texture features
• support vector machine classification

• 11 — Villes et communautés durables

• Automated Road and Building Extraction
• Remote-Sensing Image Classification
• Advanced Neural Network Applications
• building detection
• high-resolution aerial imagery
• superpixels
• texture features
• support vector machine classification
• Ocean Engineering
• Media Technology
• Computer Vision and Pattern Recognition
• Engineering
• Computer Science