Speckle Noise Reduction via Optimized Non-Local Means Filtering for High-Speed 3D Optical Metrology

This paper proposes an Optimized Non-Local Means (ONLM) algorithm for efficient speckle noise removal in fringe projection imaging. By leveraging local column pixels as a search window and differential Gaussian weighting, ONLM achieves real-time denoising while preserving image fidelity. Combined with multi-frequency phase-shift reconstruction, the method generates high-accuracy 3D point clouds. Experimental results demonstrate ONLM’s superior speed (15×faster than Fast NLM) and consistent noise suppression (PSNR~3dB). Point cloud analysis reveals sub-wavelength accuracy (2σ distance: 1.140), validating its potential for high-speed optical metrology.