Ratko Ivković, Lazar Mosurović, Radivoj Prodanović, Stefan Panić
Noise in digital images is a common artifact introduced during acquisition, processing, or transmission, often degrading visual quality and analysis accuracy. Traditional noise types, such as Gaussian, salt-and-pepper, and Poisson noise, focus on isolated pixel-level distortions and are mathematically well defined. However, real-world phenomena, like snow or rain, create structured noise that spans multiple pixels and exhibits spatial coherence. This paper introduces a novel hybrid model, S3N-AI (structured snow simulation noise via artificial intelligence), for modeling and simulating snow effects in digital images using AI-inspired parametric control. S3N-AI combines classical image noise modeling with AI-driven probabilistic rules and structural priors that emulate natural snowfall behavior. Snowflakes are generated using adaptive, depth-aware Gaussian functions guided by learned spatial distributions, enabling intensity variation, directional coherence, and perceptual realism. A density parameter enables precise control over coverage, while affine coordinate transformations simulate wind and gravity effects. S3N-AI bridges classical and AI-enhanced noise models, offering a robust simulation framework applicable in virtual environments, vision systems, and deep learning training datasets. The MATLAB-based implementation ensures reproducibility and paves the way for future integration with neural image generation techniques.