基于改进Criminisi算法的图像修复方法研究

Research on Image Restoration Method Based on Improved Criminisi Algorithm

摘要: 本文针对传统Criminisi算法在图像修复中存在的计算效率低、边缘匹配精度不足等问题，提出了一种基于传统Criminisi算法的优化方案。通过引入梯度预计算、优化局部搜索、动态置信度衰减、区块匹配加速策略，实现了修复速度与质量的协同提升。实验结果表明，本文算法的PSNR较传统Criminisi方法提升1.9 dB，改善了结构边缘的视觉连续性(SSIM提升0.028)。

Abstract: This paper addresses the issues of low computational efficiency and insufficient edge matching accuracy in the traditional Criminisi algorithm for image inpainting by proposing an optimized solution based on the conventional Criminisi algorithm. By introducing gradient precomputation, optimized local search, dynamic confidence decay, and block-matching acceleration strategies, the method achieves a synergistic improvement in both restoration speed and quality. Experimental results demonstrate that the proposed algorithm achieves a PSNR improvement of 1.9 dB compared to the traditional Criminisi method, while enhancing the visual continuity of structural edges (SSIM improvement of 0.028).