2-DIMENSIONAL NEURAL HYBRID FILTERS USING ADAPTIVE WINDOWS AND LAYERED MEDIAN FILTERS

A new 2-D neural hybrid filter that combines a layered median filter w ith adaptive-window linear filters is described. The filter can elimin ate both impulsive and. Gaussian noise and has good edge preservation characteristics, owing in particular to the 3-point layered median fil ter design. The size of the adaptive windows is maximized in regions w here the neighboring pixel values differ relatively little from that a t the working point. The design yields effective noise removal without edge degradation, and learning by the neural network allows the filte r to adapt to image characteristics.