A random-field model-based algorithm for anomalous complex image pixel detection

Random-field model-based algorithms for the detection of anomalous pixels associated with complex valued imagery may be essential to robust focus of attention, target detection, and curing. The described algorithm includes the fitting of a specific class of causal, two-dimensional autoregressive random-field models to image data over specified estimation windows, and then subsequent construction of prediction error samples over specified detection windows. Statistical testing of the calculated prediction error samples is then used to localize anomalous image pixels. Experimental results obtained from running the described algorithm on SAR (synthetic aperture radar) imagery are included.<
>