Non-Local Means Algorithm: An Adaptive Spatial Filtering Approach

The Non-Local Means algorithm is an adaptive spatial filtering method that can be viewed as an enhancement of Yaroslavsky filtering and bilateral filtering techniques. Its core principle involves estimating pixel values using similar patches within the image to reduce noise while preserving fine details. While Yaroslavsky and bilateral filters served as precursors to this approach, Non-Local Means improves filtering performance by introducing adaptive weighting based on patch similarity. The algorithm compares image patches using similarity metrics (typically Euclidean distance or normalized cross-correlation) and assigns higher weights to more similar patches during pixel averaging. This patch-based comparison enables better preservation of textures and details. Consequently, the Non-Local Means algorithm has gained widespread adoption in image processing applications, particularly in denoising and image enhancement tasks. Implementation typically involves sliding window operations with optimized search strategies for efficient patch matching.