Powers-of-2 finite-impulse-response filters in edge detection

Convenient methods based on suitably designed finite-impulse-response (FIR) or infinite-impulse-response (IIR) filters were proposed recently for identifying edges in machine vision. We show that under normal image conditions (which are satisfied virtually always in practical indoor scenes) a FIR filter with powers-of-2 coefficients is able to provide results that are equivalent to those of an 8-bit FIR filter or a 16-bit HIR filter. This permits a significant simplification in the implementation design for edge detection. Results for common two-dimensional images processed with the simplified algorithm are presented, and a comparison is made with results obtained from so-called optimal nonapproximated operators.