基于Criminisi算法的图像修复研究
Research on Image Restoration Based on Criminisi Algorithm

数字图像修复在2000年首次提出，关键应用领域包括文物保护和电影特效制作，其中Criminisi算法备受研究关注。本文针对Criminisi算法仅适用于纹理信息丰富或大区域破损图像修复的局限性展开研究，结合权重与距离参数更新Criminisi算法的优先权计算公式和匹配块匹配策略，提出一种新的Criminisi改进方法，以提高算法的适用范围和修复的图像质量。本文引入权重系数以度量优先权公式中置信度项和数据项的信息，同时引入待修复块与全局最佳匹配块间的欧氏距离作为匹配块搜索新策略，通过调节比例系数调整不同应用场景的搜索策略，以增强匹配块与待修复块的相似性。最后，本文使用PSNR作为客观评价指标，在Matlab平台完成不同图像修复任务和应用场景的修复对比实验，结果表明，本文方法不仅有效提高修复图像的平均PSNR，还能更好地适应不同修复任务及其应用场景，为数字图像修复领域的深入研究与应用提供一定价值的参考。
Digital image restoration was first proposed in 2000, and the key application fields include cultural relics protection and film special effects production, among which the Criminisi algorithm has at-tracted much research attention. Aiming at the limitation that Criminisi algorithm is only suitable for repairing damaged images with rich texture information or large areas; this paper proposes a new Criminisi improvement method by combining weight and distance parameters to update the priority calculation formula and matching block matching strategy of Criminisi algorithm to im-prove the application range of the algorithm and the quality of the restored image. In this paper, the weight coefficient is introduced to measure the information of confidence items and data items in the priority formula, and the Euclidean distance between the block to be repaired and the glob-ally best matched block is introduced as a new search strategy for matched blocks. The search strategy for different application scenarios is adjusted by adjusting the proportional coefficient to enhance the similarity between the matched block and the block to be repaired. Finally, this paper uses PSNR as an objective evaluation index to complete repair comparison experiments of different image repair tasks and application scenarios on the Matlab platform. The results show that the proposed method can not only effectively improve the average PSNR of repaired images, but also better adapt to different repair tasks and their application scenarios. It provides a valuable refer-ence for the in-depth research and application of digital image restoration.