Automated algorithms for extracting urban features from Ikonos satellite data. A case study in New York City

This paper describes development of feature extraction algorithms using spectral and spatial attributes for detecting specific urban features. Spectral and spatial heterogeneity of urban environment present challenges to their accurate detection and classification from remotely sensed data. Methods include segmenting Ikonos data, computing attributes for creating image objects, and classifying the objects with rules. Low class accuracies were reported for dark and gray roofs. By employing different segmentation scale parameters and a modified approach to feature extraction, we further improved the accuracy. In this approach new rules and attributes were applied selectively to image areas having similar or near-similar spectral and spatial characteristics that could be specified within a threshold. Results showed a remarkable improvement in the accuracy of classes with low spectral separability. We developed different algorithms using a range of spectral and spatial attributes to extract specific urban features from any mss (4m by 4m) Ikonos data. Key-Words: Ikonos, Spatial, Spectral, Image, Objects