To develop an effective algorithm for crack segmentation in Tibetan heritage mural structures, addressing challenges like poor lighting and complex backgrounds.
Developed a component tree-driven MSSR-RG algorithm based on modified single-scale Retinex (MSSR) and region growing (RG) algorithms.
Modified source code of SSR to enhance image quality under low illumination.
Established hierarchical relationships in the component tree by adjusting MSSR parameters iteratively.
Implemented a masking algorithm to optimize similarity criteria for pixel classification.
Achieved over 90% F1-score for crack segmentation through comparison with high-precision manual annotations.
Demonstrated increased efficiency in crack detection under difficult conditions, confirming the algorithm's effectiveness.