Randomize Floyd

Traditional Floyd-Steinberg (FS) error diffusion is a fundamental halftoning technique that yields high perceptual quality, but is hampered by two major drawbacks: the generation of visually intrusive “worm-like” artifacts and an inherently serial implementation that precludes efficient parallelization. This paper addresses both limitations by proposing and empirically validating a series of variations centered on randomization of the kernel weight coefficients. We develop an interactive simulation interface to systematically test the effects of stochastic elements, adaptive visual models, and parallel block processing. Our results, verified through visual comparison and quantitative spectral analysis (RAPSD and Anisotropy measures), demonstrate that introducing randomized weights successfully disrupts the deterministic error path, effectively suppress- ing low-frequency artifacts and driving the noise spectrum toward an ideal blue noise distribution. Randomization decouples pixel dependencies, enabling better-quality results even with parallel approaches like block-based process- ing. This work concludes that weight randomization is the key mechanism for achieving perceptually superior, artifact-free error diffusion while simultaneously facilitating the path toward modern, efficient parallel implementations.