Algorithm 1 Pseudocode for Procedural Snowfall Augmentation

Input: Original image $I_{orig}$ Image directory D (for batch processing) Number of snowflakes $N_{snow}$ Mask transparency ${\alpha }_{mask}$ Snow particle size range $[s_{min},s_{max}]$ Snow mask output path $M_{out}$ Output: Snow-augmented image $I_{snow}$ Step 1: Generate Snow Mask 1.Create an empty RGBA mask M with the same size as $I_{orig}$. 2. For $i=1$ to $N_{snow}$: Randomly sample position $(x_i,y_i)$ in the image. Randomly select particle type: grain or streak. If grain: – Sample radius $r_i$ from $[s_{min},s_{max}]$. – Sample alpha ${\alpha }_i$ from $[{\alpha }_{min},{\alpha }_{max}]$. – Draw an irregular filled polygon at $(x_i,y_i)$ with $r_i$ and ${\alpha }_i$ on M. If streak: – Sample length $l_i$ from $[2.5r_i,7.0r_i]$, width $w_i$ from $[0.8,1.4]$. – Sample angle ${\theta }_i$ from $[-{\displaystyle \frac{\pi }{4}},{\displaystyle \frac{\pi }{4}}]$. – Sample alpha ${\alpha }_i$ from $[{\alpha }_{min},{\alpha }_{max}]$. – Draw a line at $(x_i,y_i)$ of length $l_i$, width $w_i$, angle ${\theta }_i$, ${\alpha }_i$ on M. 3.Apply Gaussian blur (e.g., $\sigma =0.6$) to M for realism. 4.Save M as a snow mask to $M_{out}$. Step 2: Blend Snow Mask with Original Image 1. For each image $I_{orig}$ in directory D: Resize M to match $I_{orig}$ if needed. For each pixel $(x,y)$: – Compute blended pixel: $I_{snow}(x,y)=(1-{\alpha }_{mask})·I_{orig}(x,y)+{\alpha }_{mask}·M_{RGB}(x,y)$ Save $I_{snow}$ to $M_{out}$. 2.Output the snow-augmented images.