Array Generator Design for an Optical Analog-to-Digital Converter

Within the context of image processing, digital image halftoning is an important class of analog-to-digital (A/D) conversion. Halftoning can be thought of as an image compression technique whereby a continuous-tone, gray-scale image is printed or displayed using only binary-valued pixels. One method to achieve digital halftoning is error diffusion, wherein the error associated with a nonlinear quantization process is diffused within a local region. One of us (BLS) has developed a neural network architecture based on the mathematical foundation of the error diffusion algorithm that is called an error diffusion neural network [1]. The error diffusion neural network computes the halftoned image asymptotically faster than a conventional Hopfield-type neural network, provides full-rank connectivity across the entire image (other error diffusion techniques provide only local error diffusion), and, because of its parallel implementation, does not generate any artifacts commonly associated with sequential halftoning. Figure 1 is a representation of a smart-pixel-based architecture for an optical implementation of the error diffusion algorithm. The functionality required of the error diffusion neural network is implemented using gallium arsenide (GaAs) and aluminum gallium arsenide (AlGaAs) multiple quantum well modulators. The two-dimensional spatial distribution and intensity weighting required of the error diffusion filter is accomplished using a diffractive array generator. We report here on the design of the array generator.