Scholarlink

The very fast advancement of artificial intelligence (AI) in image generation has made it very difficult to differentiate AI-generated images from human generated images. This differentiation is important for digital forensics, media integrity, and intellectual property protection. In this paper, we go through various deep learning models, including EfficientNet-B0, EfficientNet-B3, and ResNet152V2, to classify AI-generated and human-generated images. Our process in the paper involves step by step fine-tuning of these architectures, using a dataset containing both human generated and AI-generated images. Final results show that transitioning from EfficientNet-B0 to ResNet152V2 improves classification performance, giving us a final validation accuracy of 98% and an AUC of 0.99. These results demonstrate the effectiveness of DL models in detecting AI-generated images and focus on the importance of continued advancements in image authenticity classification to mitigate risks associated with misinformation and digital forgery.

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

Deep Learning Based AI-Generated Image Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vishnu Shankar,
  • Asfiya Zaman,
  • Vijay Kumar Jha

摘要

The very fast advancement of artificial intelligence (AI) in image generation has made it very difficult to differentiate AI-generated images from human generated images. This differentiation is important for digital forensics, media integrity, and intellectual property protection. In this paper, we go through various deep learning models, including EfficientNet-B0, EfficientNet-B3, and ResNet152V2, to classify AI-generated and human-generated images. Our process in the paper involves step by step fine-tuning of these architectures, using a dataset containing both human generated and AI-generated images. Final results show that transitioning from EfficientNet-B0 to ResNet152V2 improves classification performance, giving us a final validation accuracy of 98% and an AUC of 0.99. These results demonstrate the effectiveness of DL models in detecting AI-generated images and focus on the importance of continued advancements in image authenticity classification to mitigate risks associated with misinformation and digital forgery.