【摘要】 由于秦简长期埋于地下，其表面受雨水侵蚀，因此在挖掘过程出现大量残简、断简。针对简牍上文字模糊不清、字体结构信息缺失的问题，文中提出一种基于改进条件扩散模型的残简文字修复方法，通过优化噪声估计网络，捕获秦简文字的全局结构信息和局部纹理细节，从而得到更好的图像修复效果。在优化网络中，首先，引入一个双信息输入的空间通道重建残差模块，提高修复的纹理清晰度；其次，引入全局上下文网络（GCNet），将局部纹理特征传播到全局，解决相隔较远文字像素间依赖关系较弱的问题，使修复后的文字结构更加完整。在自制秦简文字数据集上与其他图像修复模型进行比较，仿真结果表明：所提方法在主观视觉上达到了较好的效果，峰值信噪比和结构相似性分别达到了34.12 dB和97.8%，相比较于扩散模型Palette分别提高了1.68 dB和0.6%，修复效果满足考古工作者、文字学专家及历史学家的视觉需要。更多还原

【Abstract】 Qin bamboo slip character were buried underground for a long time, and their surfaces were eroded by rainwater, so a large number of broken bamboo slips appeared during the process of excavation. In view of the unclear characters and missing font structure information on bamboo slips, a method for repairing broken bamboo slip characters based on improved conditional diffusion model is proposed. By optimizing the noise estimation network, the global structure information and local texture details of Qin bamboo slip characters are captured, so as to obtain better image inpainting effect. In the optimized network, a spatial channel reconstruction residual module with dual information input is introduced to improve the texture clarity of the repaired image first, and then the global context network GCNet is introduced to propagate local texture features to a global range, so as to strengthen the dependency between pixels of characters far apart and make the repaired text structure complete. The comparison experiments were carried out between the proposed model and the other image repair models on the self-made Qin bamboo slip character dataset. The simulation results show that the proposed method achieves good results in subjective vision, with peak signal-to-noise ratio(PSNR) and structural similarity(SSIM) reaching 34.12 dB and 97.8%, respectively, which are 1.68 dB and 0.6% higher than those of the diffusion model Palette, respectively. To sum up, the inpainting effect can meet the visual needs of archaeologists, philology experts and historians.更多还原