(V. Spokoiny and J. Polzehl, Berlin)
 

A new type of spatial adaptive smoothing procedures, referred to as Adaptive Weights Smoothing (AWS), has been developed at WIAS. These procedures employ a structural assumption of a local constant model or of local stationarity. First applications for image reconstruction and for signal detection in functional Magnet Resonance Imaging (fMRI) are very promising. Within this application we intend to

• improve the theoretical basis of AWS
• generalize the adaptive weights smoothing approach to local parametric structural assumptions and more general classes of models, i.e. binary response models and density estimation.
• investigate the applicability on problems of image segmentation and object recognition.
• improve the approach for signal detection in fMRI, e.g. by incorporation of correlation structures and use of anatomical and physiological information.
• explore new fields of application, e.g. Synthetic Aperture Radar Imaging, segmentation problems in geology, discrimanant analysis and classification.