THE SYNTHETIC IMAGE TESTING FRAMEWORK (SITEF) FOR THE EVALUATION OF MULTI-SPECTRAL IMAGE SEGMENTATION ALGORITHMS

摘要

The segmentation stage is a key aspect of an object-based image analysis system. However, the segmentation quality is usually difficult to evaluate for satellite images. The Synthetic Image TEsting Framework (SITEF) is a tool to evaluate and compare image segmentation results. This paper presents the SITEF with an extension to model adjacency effects between neighboring parcels, using the sensor's point spread function and a grid offset. A practical application of SITEF is presented using a SPOT HRG satellite image, with 6 vegetation land cover classes identified on a mountainous area. The segmentation results were evaluated under various perspectives, including the parcel size and shape, the land cover types, the sensor grid offset and one parameter used in the segmentation algorithm.