Multi-scale contrast-to-noise ratio (MS-CNR): a novel metric for quantitative defect characterisation without manual region specification

In numerous research domains where imaging plays a pivotal role in analysing specific objects or processes, it is crucial to quantitatively evaluate the performance of acquisition systems and processing algorithms in differentiating the target from its background. This paper presents the Multi-Scale Contrast-to-Noise Ratio (<i>MS-CNR</i>) metric, a novel tool for precise defect quantification across various imaging modalities. The <i>MS-CNR</i> metric employs the Laplacian of Gaussian (LoG) operator to analyse contrast at multiple scales, allowing for effective quantitative defect characterisation without relying on predefined regions for defects or noise. Through comprehensive evaluation with synthetic and real data, the <i>MS-CNR</i> metric demonstrates a strong correlation with human visual perception and other well-established SNR metrics. It provides consistent and reproducible results, outperforming traditional SNR metrics that may be affected by specific types of noise. The <i>MS-CNR</i> metric’s robust performance and alignment with visual assessments make it a valuable addition to imaging analysis, offering a reliable and automated approach for evaluating defect visibility.在众多以成像技术为核心开展特定目标或过程分析的研究领域中，定量评估采集系统与处理算法区分目标与其背景的性能，是一项至关重要的工作。本文提出了多尺度对比度信噪比（Multi-Scale Contrast-to-Noise Ratio, MS-CNR）指标，一款可适配多种成像模态、实现精准缺陷量化的新型分析工具。该MS-CNR指标采用高斯拉普拉斯（Laplacian of Gaussian, LoG）算子在多尺度层面分析对比度，无需预先设定缺陷或噪声区域，即可完成有效的定量缺陷表征。通过合成数据与真实数据的全面评估，MS-CNR指标展现出与人类视觉感知及其他成熟信噪比（Signal-to-Noise Ratio, SNR）指标的高度相关性。其结果具备良好的一致性与可复现性，且优于易受特定类型噪声干扰的传统SNR指标。MS-CNR指标的优异鲁棒性与视觉评估适配性，使其成为成像分析领域极具价值的补充手段，可为缺陷可见性评估提供可靠且自动化的解决方案。