<![CDATA[This study demonstrates the comparison of five well-known gradient operators —Sobel, Prewitt, Roberts, Scharr, and Laplacian of Gaussian (LoG)—using the classical technique of Otsu thresholding to detect microscopic porosities in SMAW welds of ship structures. This was done using 33 images of micrographs taken at 25x magnification, with all images preprocessed using standardization techniques (grayscale conversion, Gaussian filtering, and normalization), and the results were compared using six different performance evaluation criteria —Precision, Recall, F-Measure, Peak Signal-to-Noise Ratio (PSNR), Edge Co-occurrence Matrix (ECM), and Matthews Correlation Coefficient (MCC). The results indicate that the Sobel operator produced the largest PSNR value (≈42.5 dB) indicating that the Sobel operator provided the greatest fidelity in image quality. Additionally, the Scharr operator had the largest MCC (0.94) and ECM correlation (0.99), indicating that it provided the greatest reliability and coherence in texture. The Roberts operator produced the largest F-measure (0.23), indicating that it provided a good balance between the two evaluation metrics. In summary, the Scharr operator demonstrated the most consistent and interpretable performance among the three performance evaluation metrics, providing both the best visual fidelity and structural texture stability. Therefore, the proposed methodology represents a reproducible and computationally-efficient reference for automated microscopic inspections and smart NDT in maritime welding applications.]]>
