文字部件分割方法

Text Component Segmentation Method

残损古碑文样本少且语义多样,文字部件检测和分割是残损碑文从自身空间结构复原必不可少的环节。提出了对残损碑文部件进行检测和分割的方法。基于SOLOv2的改进模型,通过层级融合模块融合多层特征的语义信息和细粒度特征,提升特征表征能力;通过边界增强模块引入边缘先验融合预分割特征,增强掩码边界精度。结果表明,所提出模型的分割精度提升1.4%~16.2%,验证了该模型的有效性。

Damaged ancient inscriptions have few samples and diverse semantics,and text component detection and segmentation are essential steps in restoring damaged inscriptions from their own spatial structure.A method is proposed to segment damaged inscriptions.We present an improved model based on SOLOv2.It integrates semantic information and finegrained features of multilayer features through a hierarchical information module(HIM)to enhance feature representation capabilities,and introduces edge prior fusing presegmentation features by the boundary enhancement module(BEM)to enhance the accuracy of mask boundaries.The results show that the proposed model improves the segmentation accuracy by 14%to 162%,and further verifies the effectiveness of the proposed model in the text components segmentation of the inscriptions.