A Novel Detail-Enhancement Method for Industrial Digital Radiography via Gaussian-Free Multiscale Laplacian Adaptive Fusion

Abstract

This study introduces a concise framework for detail enhancement in industrial digital radiography based on the mathematical integration of logarithmic transformation and multiscale Laplacian analysis. The proposed method utilizes multiscale adaptive pixel-level fusion with hyperbolic tangent-based coefficients to preserve microscale defects while enhancing subtle features throughout the dynamic range. Quantitative evaluations of diverse industrial welds, including ship plates, boilers, and oil pipelines, demonstrated substantial improvements. In oil pipeline weld inspections, the Peak Signal-to-Noise Ratio of the method based on Histogram Equalization increased by 133.82%, whereas the Structural Similarity Index Measure and Spatial Frequency metrics exhibited gains of up to 127.27% and 85.81%, respectively. The framework's consistent, albeit moderate, performance gains over state-of-the-art deep learning methods across all benchmarks confirm its value not only as a robust and widely applicable tool but also as a superior preprocessing or integrated solution within nondestructive testing pipelines.