Performance Assessment of Automated Feature Extraction Tools on High Resolution Imagery

The enormous increase in the volume of remotely sensed data that is being acquired by an ever-growing number of earth observation sensors, led to the situation where the image analysts are at risk to be overwhelmed by the number of images to process. A way to cope with this situation is to use automated feature extraction (AFE) tools to help the image analysts in the extraction of geospatial features, and address specific tasks such as object detection for change detection purposes. However, it is not clear which AFE tools perform better against others, and on what scenarios. This paper presents an assessment of the performance of three AFE tools to extract man-made objects for change detection on high resolution airborne and spaceborne electro-optic imagery: Feature Analyst (Visual Learning Systems Inc.), eCognition (Definiens AG), and Genie Pro (U.S. Los Alamos National Laboratory). These tools all intend to assist the image analysts in extracting relevant image features using dissimilar feature extraction algorithms. Feature Analyst employs machine learning techniques to extract user-defined features; eCognition is an object-oriented classification package using a fuzzy logic scheme; Genie Pro is an evolutionary computation software system assembling image processing tools from a collection of image operators by means of genetic algorithms. The performance of these automated feature extraction software packages in extracting relevant geospatial features on high resolution remotely sensed images is assessed, together with their use for change detection applications.