Look-up tables are a simple and an efficient means of providing neighborhood transformations in low level image processing architecture; however, they can also form a significant portion of the hardware cost in parallel systems.This paper describes a look-up table structure for low level machine vision applications in which only a restricted set of 3 × 3 neighborhood transformations are necessary and where the size, cost and real-time operation of the processor are additional constraints. The structure is based upon a novel two stage cascaded memory network which acts as the look-up table to implement binary neighborhood transformations. The structure provides sufficient realizations for many low level operations required in machine vision applications, but requires only one-sixth of the gates that would be necessary to provide complete cover.The paper presents a description of the structure, proofs for its realizability and hardware costs based upon simulated designs.