Automated Stratification and Noise-based Dynamic Parameter Tuning for Robust Extraction of Inscribed Text from Ancient Stone Inscriptions

Ancient stone inscriptions constitute a significant archive of heritage, language, and cultural diversity, but their preservation and interpretation are challenging due to surface degradation, deterioration, stone background, texture, brightness, illumination, inscribed depth, type of palaeography and the like. As a result of an insightful review of prior art, an inclusive and unified solution to address variation in imaging features and dynamic parameter tuning in pre-processing is required. In this study, a two-phase framework is proposed. In Phase I, inscription images are classified using Elbow-based K-Means clustering with features. This data-driven stratification facilitates the systematic evaluation of heterogeneous image features, ensuring that associated images are processed under similar enhancing conditions. In phase II the Dynamically Parameterized Image Pre-processing Pipeline (DPIPP) is used, that dynamically tunes fundamental pre-processing parameters based on noise estimation. This noise-adaptive mechanism automatically adjusts denoising strength and binarization thresholds for each image, resulting in optimal enhancement regardless of illumination, texture, or surface degradation. A quantitative evaluation using image fidelity metrics (PSNR, MSE, SSIM, Entropy, and SD) and segmentation-based metrics (Precision, Recall, Dice, Detection Accuracy) demonstrates that the proposed method outperforms static pre-processing approaches in terms of contrast enhancement, background reduction, and text readability. This study builds a strong and scalable pre-processing basis for digital epigraphy by combining feature-driven clustering with noise-adaptive parameterization, improving the reliability of downstream segmentation, recognition, and heritage preservation.