Standard image processing techniques are not applicable to radar images because of the coherent nature of the sensor. Therefore there is a need to develop preprocessing techniques for radar images which will then allow these standard methods to be applied. A random field model for radar image data is developed. This model describes the image data as the result of a multiplicative-convolved process. Standard techniques, those based on additive noise and homomorphic processing are not directly applicable to this class of sensor data. Therefore, a minimum mean square error (MMSE) filter was designed to treat this class of sensor data. The resulting filter was implemented in an adaptive format to account for changes in local statistics and edges. A radar image processing technique which provides the MMSE estimate inside homogeneous areas and tends to preserve edge structure was the result of this study. Digitally correlated SEASAT-A synthetic aperture radar (SAR) imagery was used to test the technique.

Date of this Version