Scholarlink

<p>In the context of the digital era’s unprecedented expansion, the proliferation of disinformation and fabricated narratives has emerged as a critical societal challenge. This trend not only weakens the trust in mainstream media but also exposes internet users to substantial risks, as they often unknowingly propagate misleading content. The emergence of diverse media formats, including textual fabrications, doctored images, and manipulated videos, worsens the challenge of distinguishing fact from fiction. Despite coordinated efforts within the computing research community, detecting and generating deceptive content remains a challenging task. This study concentrates on the generation of fake images, a particularly deceptive variant of misinformation. Introducing the Cycle Generative Adversarial Fake pattern transfer (FAKE-CGAN) deep learning model, we attempt to craft synthetic fake images that closely mirror authentic counterparts. Leveraging the capabilities of CycleGAN, our methodology navigates unpaired image-to-image translation, allowing FAKE-CGAN to learn mappings between datasets without explicit pairings, thereby amplifying its versatility. Rigorous evaluations, encompassing adversarial loss metrics such as Generator adversarial loss, Discriminator adversarial loss, Cycle consistency loss, and Identity loss, enable a detailed assessment of simulated patterns’ accuracy and their alignment with real scenes. Additionally, multiple standard generative evaluation metrics are utilized to quantitatively validate the improved performance of the proposed Fake-CGAN model. This research contributes to advancing the frontier of fake image generation through a novel framework, elucidating the intricate complexities of deceptive visual content, and providing a foundational pathway for future research endeavours.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Pattern transfer based photorealistic synthetic fake image generation using cycle generative adversarial networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月24日

查看全文
  • Anuja Arora,
  • Jasir Mohammad Zaeem,
  • Vibhor Garg,
  • Kirti Aggarwal,
  • Xiao-Zhi Gao

摘要

In the context of the digital era’s unprecedented expansion, the proliferation of disinformation and fabricated narratives has emerged as a critical societal challenge. This trend not only weakens the trust in mainstream media but also exposes internet users to substantial risks, as they often unknowingly propagate misleading content. The emergence of diverse media formats, including textual fabrications, doctored images, and manipulated videos, worsens the challenge of distinguishing fact from fiction. Despite coordinated efforts within the computing research community, detecting and generating deceptive content remains a challenging task. This study concentrates on the generation of fake images, a particularly deceptive variant of misinformation. Introducing the Cycle Generative Adversarial Fake pattern transfer (FAKE-CGAN) deep learning model, we attempt to craft synthetic fake images that closely mirror authentic counterparts. Leveraging the capabilities of CycleGAN, our methodology navigates unpaired image-to-image translation, allowing FAKE-CGAN to learn mappings between datasets without explicit pairings, thereby amplifying its versatility. Rigorous evaluations, encompassing adversarial loss metrics such as Generator adversarial loss, Discriminator adversarial loss, Cycle consistency loss, and Identity loss, enable a detailed assessment of simulated patterns’ accuracy and their alignment with real scenes. Additionally, multiple standard generative evaluation metrics are utilized to quantitatively validate the improved performance of the proposed Fake-CGAN model. This research contributes to advancing the frontier of fake image generation through a novel framework, elucidating the intricate complexities of deceptive visual content, and providing a foundational pathway for future research endeavours.