Image denoising method combining wavelet denoising and improved non-local mean algorithm

Traditional wavelet threshold denoising fails to fit noise across decomposition layers and has biased discontinuous threshold functions. This paper presents an efficient composite denoising algorithm for complex noise. It first designs a multi-scale dynamic thresholding scheme with a continuous low error threshold function to address wavelet defects. Then it improves non-local means via block multi-scale decomposition and fast-decaying weights for better block matching and weight assignment. A composite denoising framework with progressive processing and coordinated parameters is built. Tests on Set12 and BSD68 datasets yield 0.90 precision, 0.95 recall, up to 29.87dB PSNR, 0.95 SSIM, ∼1.0s runtime, and near-0.01 error metrics. The method performs well on images with dense edges, intricate textures and missing data.