Non-Local Means Denoising Visualization

How Non-Local Means Works

Non-local means denoising is a powerful image denoising algorithm that leverages self-similarity within images. Here's what's happening in the visualization:

1. Reference Patch: For each pixel (x,y) in the image, a patch centered at that pixel is extracted (green square).
2. Search Region: The algorithm searches within a neighborhood around the pixel (blue square).
3. Patch Comparison: Each patch in the search region is compared to the reference patch.
4. Similarity Weights: Weights are assigned based on patch similarity, with more similar patches receiving higher weights.
5. Weighted Average: The final denoised value for the pixel is a weighted average of all center pixels from the compared patches.

The sliders control:

• Patch Size (s): The size of square patches to compare (larger patches capture more structure but may blur details).
• Search Distance (d): How far from the current pixel to search for similar patches.
• Filter Parameter (h): Controls the decay of the weighting function (smaller values preserve more detail but may retain noise).
• Variance (var): Related to the expected noise level in the image (higher values for noisier images).