Enhancing subtle details in industrial radiography: a physics-inspired framework integrating anisotropic crystal growth kinetics with multiscale fluid dynamics
Ochiq kirish: yoʻq
Annotatsiya
Abstract In 16-bit industrial radiography, existing enhancement techniques suffer from noise amplification, artifacts, poor interpretability, and high data requirements. To address these limitations, this study proposes a training-free and interpretable framework that enhances subtle details while preserving structural fidelity. This methodology reconceptualizes image enhancement as the simulated evolution of a virtual material field.It integrates anisotropic crystal growth kinetics based on the Burton-Cabrera-Frank model with multiscale fluid dynamics derived from Perona-Malik anisotropic diffusion, establishing a dual-physics analogy for local edge sharpening and global structure preservation. A locally adaptive crystal growth field constructed from pixel-level lattice energy and an energy-weighted gradient guides a multiscale pyramid decomposition that selectively amplifies the high-frequency details while suppressing noise. Experiments on industrial radiographic datasets using the recommended global enhancement strength (η = 7) demonstrated an average structural similarity index of 0.9726, a peak signal-to-noise ratio of 30.15 dB, and a 126.52% relative improvement in spatial frequency, consistently outperforming established benchmarks. This paradigm offers a practical alternative to manual windowing for human and automated inspections, establishing a methodological bridge between physical simulations and computational imaging for high-stakes, nondestructive evaluations. The code and data are archived with a persistent
Hali tarjima qilinmagan