Image processing often involves operations using `neighbor' pixels. This paper combines the usual definition of 4 or 8 connected neighbors with image information to produce local neighbor definitions that are signal dependent. These generalized neighbors, G-neighbors, can be used for a variety of image processing tasks. The paper examines their use for detail preserving smoothing and morphology. The simple/local nature of G-neighbor definitions make them ideal for implementation on low-level pixel parallel hardware. A near real-time parallel implementation of the G-neighbor computation, including G-neighbor-based detail preserving smoothing and G-neighbor morphology, is discussed. The paper also provides a qualitative comparison of G-neighbor-based algorithms to previous work.