Answers are grounded in this paper and cite their sources.
In medical imaging and industrial non-destructive testing (NDT), preserving structural features while suppressing noise presents a fundamental trade-off. To overcome this limitation, this study proposes an adaptive image processing (AIP) technique utilizing a gradient-based delta-map (\(\:δ\:\)-map) that generates pixel-wise weights based on the local gradient magnitudes. The proposed method quantifies the structural salience of the input image to generate a \(\:δ\:\)-map and applies it to a residual-based enhancement process, thereby selectively suppressing noise in homogeneous regions while enhancing sharpness in edge regions. Experimental results on printed circuit board (PCB) X-ray images demonstrated that the proposed method yielded superior visual quality compared to conventional filters with globally fixed parameters (e.g., Gaussian, unsharp masking, bilateral) and commercial algorithms (MUSICA). Consistent performance improvements were confirmed through quantitative metrics (SSIM, PSNR, LPIPS) and no-reference image quality assessments (BRISQUE, NIQE). Furthermore, analysis of the modulation transfer function (MTF), noise power spectrum (NPS), and noise-equivalent quanta (NEQ) verified that the proposed method optimizes signal transfer characteristics across a wide frequency range while minimizing spatial resolution degradation. Unlike deep learning approaches, this method offers high computational efficiency and algorithmic interpretability, providing a practical solution immediately applicable to real-time X-ray inspection systems.
No takes yet. Share an insight, caveat, or question.
Hwang et al. (2026) studied this question.