A novel approach to mural enhancement using MSR CAB and lacuna extraction from ancient mural paintings using random forest

Ancient mural paintings, such as those in Bey's Palace, Algeria, are vulnerable to deterioration from environmental and human factors, with lacunae areas of paint layer loss serving as key damage indicators. Existing detection methods, including resource-intensive deep learning and traditional statistical approaches, are limited by strict assumptions and high computational demands, in addition to the small datasets typically available in heritage contexts, thus making low-resource and interpretable approaches essential. This paper introduces a novel, automated lacuna extraction method using a Random Forest classifier applied to RGB images of ancient murals. The process begins with Multi-Scale Retinex with Contrast-Limited Adaptive Histogram Equalization (MSR-CAB), Adaptive Blending, and Bilateral Filtering. This new preprocessing method enhances image quality by addressing low-light conditions, dust obscurities, and uneven illumination while preserving fine details and historical authenticity. The Random Forest classifier is a non-parametric ensemble technique that, learns decision boundaries from expert-defined regions without distributional assumptions. Feature extraction integrates RGB intensities, local variance, and gradient magnitude, while a cost-sensitive Random Forest model handles class imbalance for robust pixel-wise classification. Experimental results demonstrate high accuracy, precision, and recall, highlighting the method's potential to support heritage institutions in documentation, monitoring, and conservation planning.