A Mask-Fusion Detection Method for Character Integrity Preservation and Adhesion Segmentation in Jiandu Manuscripts

Jiandu served as the primary medium for recording important information in ancient China, and the analysis and interpretation of their textual content contribute to a deeper understanding of history and the preservation of cultural heritage. Jiandu text detection is a prerequisite for subsequent text recognition and semantic analysis. To address challenges such as surface aging, irregular character scales, dense character arrangement, and intersecting character adhesion in Jiandu images, this paper proposes a Jiandu text detection method based on character-level region mask fusion and adaptive segmentation, aiming to better preserve character structures and effectively segment adhered characters. In the automatic segmentation stage, a pixel feature analysis and overlap suppression mechanism is introduced to reduce background noise and invalid regions. In the character region processing stage, mask clustering and region fusion strategies are employed to alleviate the effects of inter-character occlusion and adhesion. Subsequently, an adaptive segmentation method is applied to achieve accurate segmentation of individual characters from complex adhered character regions. Experimental results demonstrate that the proposed method achieves significant improvements in character structure preservation and adhered character segmentation, effectively enhancing the overall performance of Jiandu text detection.