摘要
为了解决文本透射图像复原方法中实际成对图像数据集缺乏而导致的去透射效果欠佳的问题,本文提出了一种基于CycleGAN的透射文本图像复原方法。利用CycleGAN的约束转移学习能力完成非成对图像的复原任务;考虑到实际透射文本图像的复杂性,结合人类视觉特征,在CycleGAN中融入注意力机制,非均匀地处理不同文本特征。实验结果表明,所提方法在公共图像数据集以及真实图像数据上都取得了较好的透射图像恢复结果。
To solve the problem of poor de bleed-through effect caused by the lack of paired image datasets in text bleed-through image restoration methods,a bleed-through text image restoration method based on CycleGAN is proposed.Utilize the constraint transfer learning ability of CycleGAN to complete the restoration task of non-paired images.Considering the complexity of actual bleed-through text images,combined with human visual features,attention mechanisms are incorporated into CycleGAN to nonuniformly process different text features.The experimental results show that the proposed method achieves good bleed-through image restoration results on both public image datasets and real image data.
作者
马咏莉
李嘉培
MA Yongli;LI Jiapei(College of Information Engineering,Zhengzhou University of Science and Technology,Zhengzhou 450064,China)
出处
《智能计算机与应用》
2024年第7期211-215,共5页
Intelligent Computer and Applications
