Research project T4/DD/08 (Research action T4)
The localisation and identification of significant changes in temporal image sequences is important for the exploitation of satellite imagery for monitoring applications. There is an increasing demand for sophisticated automatic or semi-automatic image interpretation tools, due to the increasing availability of satellite images with high temporal and/or spatial resolution.
Current techniques for the detection of changes in temporal image sequences are based on image differences and are consequently very sensitive to registration errors and photometric conditions. Even if techniques could be developed to cancel the effects, changes, of which the significance could only be judged by experts with sufficient domain knowledge, would still be detected.
The proposed project aims at the development of more advanced techniques for change detection by making use of semantic information in the form of reference images. Instead of detecting the changes at the level of the individual pixels is proposed to detect changes at higher semantic level by using model based segmentation techniques.
Simple but effective structural pattern recognition methods which exploit local context information can be used to detect changes. Methods for the automatic construction of the models of the context will be developed.
The research will focus on monitoring using temporal SAR-satellite imagery. The specific speckle noise is problematic for change detection methods based on image differences. Segmentation methods that exploit textural homogeneity will be studied.