查詢結果分析
來源資料
頁籤選單縮合
| 題 名 | 基於深度學習的文件浮水印去除技術=Document Watermark Removal Techniques Based on Deep Learning |
|---|---|
| 作 者 | 蔡建戊; 徐泓緯; | 書刊名 | 資訊電子學刊 |
| 卷 期 | 12:2 2025.01[民114.01] |
| 頁 次 | 頁59-84 |
| 分類號 | 312.837 |
| 關鍵詞 | 浮水印去除; 條件式生成對抗網路; 影像修復; YOLOv4; Watermark removal; Conditional generative adversarial network; Image restoration; |
| 語 文 | 中文(Chinese) |
| 中文摘要 | 浮水印對於創作者來說是一種在影像中加入附加資訊去辨識和保護其自身財產權的做法,但它確實會對影像造成某種程度上的改變,因此本研究將浮水印的移除視作一種影像修復,去探討如何使用深度學習有效地去除浮水印並且恢復原始圖像資訊。在運用深度學習做影像修復的研究領域中,常見的影像修復有卷積神經網路(CNN)、生成對抗網路(GAN)、自編碼器(Autoencoders)等,而在本研究中結合了YOLOv4(You Only Look Once)的神經網路有效且快速地進行浮水印的偵測,再利用生成式影像修復網路WDNet(Watermark De-Noising Network)來加速並完成浮水印去除之影像修復任務,透過這兩種深度學習技術的結合,並將YOLOv4的邊界框進行擴大,以達到更完整且快速的文件浮水印修復之效果。所提方法不僅能準確且快速偵測出文件浮水印所在位置,同時在影像修復上亦可達到明顯有效的修復效果,並在PSNR達到45.7621db和SSIM上達到0.9842的良好結果,證明所提方法能夠有效地進行浮水印去除的影像修復及還原。 |
| 英文摘要 | Watermarks are a method used by creators to embed additional information in images to identify and protect their intellectual property. However, they do alter the image to some extent. Therefore, this study treats watermark removal as a form of image restoration, exploring how deep learning can be effectively used to remove watermarks and restore the original image information. In the field of image restoration using deep learning, common methods include Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), and Autoencoders. In this study, the YOLOv4 (You Only Look Once) neural network is combined for efficient and rapid watermark detection, and the generative image restoration network WDNet (Watermark De-Noising Network) is used to accelerate and complete the task of watermark removal and image restoration. By integrating these two deep learning techniques and expanding the bounding boxes from YOLOv4, a more comprehensive and rapid restoration of document watermarks is achieved. The proposed method not only accurately and quickly detects the location of document watermarks but also achieves significantly effective restoration results, with a PSNR of 45.7621 dB and an SSIM of 0.9842, demonstrating that the method is capable of effectively performing watermark removal and image restoration. |
本系統中英文摘要資訊取自各篇刊載內容。