Scholarlink

<p>Indian murals are valuable cultural assets but are at risk of natural aging, human vandalism, and environmental degradation. Manual restoration is often time-consuming and prone to error. To address these issues, this study proposes an image inpainting framework featuring a novel dynamic mask generation method capable of simulating diverse and realistic damage patterns. The framework also utilizes k-mean clustering algorithm to group visually similar murals to enforce specialized training. The study features three inpainting models to restore degraded murals based on the proposed masks: Partial Convolution, Deepfillv2, and Aggregated Contextual Transformations generative adversarial network. Further inpainting improvement is achieved through a weighted average technique that combines outputs from different models to form a single mask. The proposed method achieves a Structural Similarity Index of 94.88%, demonstrating its effectiveness in restoring damaged mural images with high visual accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Reconstructing degraded areas of old Indian wall paintings through image inpainting

  • 期刊论文

  • 发表日期：

    2025年10月7日

查看全文
  • Anshul Kumar Yadav,
  • Parth Naik,
  • Dhiraj

摘要

Indian murals are valuable cultural assets but are at risk of natural aging, human vandalism, and environmental degradation. Manual restoration is often time-consuming and prone to error. To address these issues, this study proposes an image inpainting framework featuring a novel dynamic mask generation method capable of simulating diverse and realistic damage patterns. The framework also utilizes k-mean clustering algorithm to group visually similar murals to enforce specialized training. The study features three inpainting models to restore degraded murals based on the proposed masks: Partial Convolution, Deepfillv2, and Aggregated Contextual Transformations generative adversarial network. Further inpainting improvement is achieved through a weighted average technique that combines outputs from different models to form a single mask. The proposed method achieves a Structural Similarity Index of 94.88%, demonstrating its effectiveness in restoring damaged mural images with high visual accuracy.