Stationary and nonstationary spatial domain Metz filtering

Stationary and nonstationary finite-impulse-response (FIR) implementations of the count-dependent Metz filter were investigated in this study. Filter size was observed to be an important variable controlling image quality. For Metz filtering of 128 X 128 pixel images at least a 15 X 15 term FIR filter was deemed necessary. By using an algorithm which selected between a set of preformed FIR filters based on pixel count, a nonstationary FIR implementation of the Metz filter was developed which required very little increase in execution time to stationary filtering. In a limited comparison of 'tumor' detection with stationary and nonstationary FIR filtering all of the Metz filtering techniques showed a significant improvement in detection when compared to the unprocessed images. However, no significant difference was observed between the stationary and nonstationary Metz filtering techniques. Thus, for Metz filters optimized solely on the basis of count, nonstationary FIR filtering does not seem to offer an advantage when compared to stationary filtering.