Component tree-driven MSSR-RG algorithm for crack segmentation in typical Tibetan heritage mural structures

Significant developments have been made in crack segmentation tasks within the single-context of modern architectural structures. However, when confronted with the challenges posed by more complex backgrounds and poor lighting conditions, the precision of crack segmentation technologies remains in a continual state of exploration and refinement. Consequently, this paper puts forward a component tree-driven MSSR-RG algorithm originally, primarily based on modified single-scale Retinex (MSSR) and region growing algorithms (CT-MSSR-RG). Initially, the source code of the SSR algorithm was modified to preserve the logarithmic-domain distribution of image data, thereby compensating for insufficient illumination. Subsequently, the vertical hierarchical relationships within the component tree (CT) are constructed by iteratively adjusting the ksize parameter of the MSSR algorithm. Simultaneously, the horizontal component relationships are established through MSSR, threshold segmentation (TS) and RG algorithms. Finally, a masking algorithm is introduced, reformulating the similarity criterion of the RG algorithm (the computations for determining whether pixels belong to the crack)—from function-based calculations to image processing approaches—to accelerate the growth of the CT. Deployed for crack segmentation in Tibetan traditional mural monitoring, a representative scenario for heritage preservation monitoring challenges, the method achieves over 90% F1-score by systematically comparing the segmentation results purely derived from intrinsic image features with high-precision manual pixel-level annotations, confirming its efficiency of crack segmentation under strong interference conditions.