Deep Learning-based Text-in-Image Watermarking
Presenter Type
UNO Graduate Student (Masters)
Major/Field of Study
Computer Science
Advisor Information
Dr. Xin Zhong
Location
CEC RM #201/205/209
Presentation Type
Poster
Start Date
22-3-2024 2:30 PM
End Date
22-3-2024 3:45 PM
Abstract
In the era of digital age, ensuring the authenticity of digital content has gained paramount of importance. To address this, this paper introduces a new technique of text-in-image watermarking. This method ingeniously embeds and extracts text watermarks within a cover image, providing a covert means of safeguarding text and image information. Our work proposes a deep learning based approach that uses Transformer [1] to learn the embedding of the text and embed it to the image using Vision Transformer [2]. This method marks a first in the field, uniquely adapting and modifying the standard Transformer [1] and [2] outputs, setting a new precedent in the application of hiding text in images and protecting both text and images. The evaluation of our proposed method involved subjecting it to various common image noises, which demonstrated its robustness and ability to resist attacks while preserving the quality of the image. Additionally, the study offers valuable insights into how various elements, like the size of the watermark, the dimensions of the cover image, and the redundancy in watermark embedding, can influence the overall performance of the watermarking process.
Deep Learning-based Text-in-Image Watermarking
CEC RM #201/205/209
In the era of digital age, ensuring the authenticity of digital content has gained paramount of importance. To address this, this paper introduces a new technique of text-in-image watermarking. This method ingeniously embeds and extracts text watermarks within a cover image, providing a covert means of safeguarding text and image information. Our work proposes a deep learning based approach that uses Transformer [1] to learn the embedding of the text and embed it to the image using Vision Transformer [2]. This method marks a first in the field, uniquely adapting and modifying the standard Transformer [1] and [2] outputs, setting a new precedent in the application of hiding text in images and protecting both text and images. The evaluation of our proposed method involved subjecting it to various common image noises, which demonstrated its robustness and ability to resist attacks while preserving the quality of the image. Additionally, the study offers valuable insights into how various elements, like the size of the watermark, the dimensions of the cover image, and the redundancy in watermark embedding, can influence the overall performance of the watermarking process.