Quality preservation in image denoising through texture-based segmentation and adaptive diffusion filtering for enhanced edge and detail retention
Lingling Mei · The Imaging Science Journal · 2025
The paper proposes an innovative adaptive noise reduction scheme that reflects the perceptual response of the human visual system by detecting its sensitivity changes across different texture areas. The method begins with entropy segmentation based on entropy to separate the image into smooth and dense-textured areas through Pixon creation and fuzzy grouping. Smoothness-adaptive and region-specific filtering is utilized subsequently, with aggressive noise reduction over smooth areas and conservative filtering for textured regions in order to recover edges and microscopic details. By avoiding excessive computation in busy areas, artifacts are reduced and perceptual fidelity is preserved. Experimental results on both benchmark sets (CSIQ and IVC), as well as authentic noisy images, show comparable improvements in both subjective quality as well as objective measures (PSNR, SSIM, FSIM, VIF, and MSER-SSIM) compared to other state-of-the-art approaches. The model’s adaptability, generalizability, and alignment with human perception ensure efficient denoising across varied conditions.