Researchers in image processing have long recognized the importance of modeling the human observer. Although a full human vision model remains elusive, orientation detection, one of the key components of human vision, can be directly incorporated into a variety of image processing algorithms. Orientation detection also provides cues that allow an algorithm to adapt to inhomogeneities in images. In this paper we show how the, a new non-linear dynamical system, can easily incorporate orientation sensitivity for two different types of problems. First, simultaneous adaptive filtering and non-linear restoration is illustrated for fingerprint enhancement. Second, constrained non-linear optimization is illustrated for halftoning in a ``hand-drawn'' style.