Advanced Hybrid Denoising Approach Using Sliced Ridgelet Transform for Image Denoising and Signal Enhancement in Medical and Industrial Domains.

Effective image denoising is critical for ensuring high-fidelity results in medical diagnostics and industrial inspection. This paper presents SRT-Net, a novel two-phase hybrid denoising framework that synergizes the mathematical strengths of the Sliced Ridgelet Transform (SRT) with the contextual power of deep learning. Phase I employs the SRT, bit-plane decomposition, and adaptive thresholding to convert a noisy image into a sparse, structurally-preserved representation. This output serves as an optimized input for Phase II, an attention-based Residual U-Net that refines textures and restores perceptual quality. Evaluated on diverse medical and industrial datasets, our framework demonstrates state-of-the-art performance, achieving a significant PSNR improvement of up to [e.g., 2.0 dB] over leading methods like BM3D. The results confirm that this hybrid, two-phase approach effectively suppresses complex noise while preserving the fine, directional details essential for accurate analysis.