| 摘 要: 为解决修复过程中壁画结构不完整与语义信息缺失等问题，提出一种基于多模态渐进融合的数字多媒体壁画修复方法。首先，构建视觉与文本的双模态特征提取网络，以获取视觉结构与语义描述信息。然后，引入信息瓶颈机制对跨模态特征进行压缩与对齐，有效增强关键语义表达。同时，设计渐进式多阶段修复策略，通过粗粒度结构重建与细粒度纹理优化协同，实现壁画全局结构一致性与局部细节真实性的统一。在敦煌壁画数据集上的结果表明，在多种损坏设置下均优于现有先进方法，为数字文化遗产保护提供有效技术路径。 |
| 关键词: 数字多媒体 多模态学习 壁画修复 信息瓶颈 |
|
中图分类号:
文献标识码:
|
|
| A Digital Multimedia Mural Restoration Method Based on Progressive Multimodal Fusion |
|
Li mo, Li nengneng, Jiao Guodong
|
Weifang Vocational College
|
| Abstract: To address the issues of incomplete mural structure and missing semantic information during the restoration process, this paper proposes a digital multimedia mural restoration method based on progressive multimodal fusion. First, a dual-modality feature extraction network is constructed to capture both visual structural information and textual semantic descriptions. Then, an information bottleneck mechanism is introduced to compress and align cross-modal features, effectively enhancing key semantic representations while suppressing redundant information. Furthermore, a progressive multi-stage restoration strategy is designed, which jointly performs coarse-grained structural reconstruction and fine-grained texture refinement, achieving consistency of global structure and realism of local details. Results on the Dunhuang mural dataset demonstrate that the proposed method consistently outperforms state-of-the-art approaches under various degradation settings, providing an effective solution for digital cultural heritage preservation. |
| Keywords: Digital Multimedia Multimodal Learning Mural Restoration Information Bottleneck |
