Improved Retinex model for enhancement and denoising of fringe patterns with non-uniform illumination

To address phase demodulation challenges in electronic speckle pattern interferometry (ESPI) fringe patterns from non-uniform illumination and heavy noise in dark regions, an improved Retinex-based denoising algorithm is proposed. First, nonlinear gamma correction is applied to the hue, saturation, and value (HSV) value channel to adaptively compress the dynamic range and improve dark-area visibility. Next, block-matching and 3D filtering (BM3D) denoises the reflectance component, while a fractional-order directional-total variation (TV) term, solved in the fast Fourier transform (FFT) domain, suppresses speckle noise while preserving sharp fringe edges. Subsequently, a Retinex self-guided mechanism performs secondary structure fidelity on illumination-reflectance components to enhance high-frequency fringe contrast. Finally, post-processing modules-multi-scale contrast limited adaptive histogram equalization (CLAHE), edge-mask sharpening, dark-zone lifting, and skin-tone saturation protection-improve visual clarity and phase demodulation reliability. The proposed method was evaluated on three simulated ESPI patterns with non-uniform illumination and five experimentally acquired low-contrast patterns, and compared with three state-of-the-art methods.