Scholarlink

<p>Over the last few years, spread of DeepFake content has become major issue on digital platforms, calling for strong detection systems. This research investigates and compares the performance of machine learning (ML) techniques and the deep learning (DL) techniques for DeepFake image categorization. The ML technologies like SVM, Random Forest, and ensemble models were utilized based on hand-designed features like GLCM and HOG, whereas DL techniques utilized strong architectures like EfficientNet-B3 and InceptionV3. The research also investigates effect of datasets of different size on model’s performance in order to analyze scalability and generalization better. Experimental findings exhibit the higher performance of DL models in accuracy, with EfficientNet-B3 obtaining the highest value of 97.95%, while ML models provided computational efficiency and interpretability. This work adds a comparative view on the practical merits and demerits of both paradigms on the problem of DeepFake detection. In addition to benchmark datasets, the model was further evaluated on AI-generated images from ChatGPT and Gemini, successfully detecting them as fake, demonstrating strong generalization.</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

An enhanced deep learning framework for DeepFake detection using EfficientNet-B3 comparative evaluation of deep and machine learning techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月10日

查看全文
  • B. G. Deepa,
  • C. K. Lokesh,
  • D. Umamaheswari,
  • B. Ayshwarya,
  • P. V. Yethish,
  • K. P. Suhaas

摘要

Over the last few years, spread of DeepFake content has become major issue on digital platforms, calling for strong detection systems. This research investigates and compares the performance of machine learning (ML) techniques and the deep learning (DL) techniques for DeepFake image categorization. The ML technologies like SVM, Random Forest, and ensemble models were utilized based on hand-designed features like GLCM and HOG, whereas DL techniques utilized strong architectures like EfficientNet-B3 and InceptionV3. The research also investigates effect of datasets of different size on model’s performance in order to analyze scalability and generalization better. Experimental findings exhibit the higher performance of DL models in accuracy, with EfficientNet-B3 obtaining the highest value of 97.95%, while ML models provided computational efficiency and interpretability. This work adds a comparative view on the practical merits and demerits of both paradigms on the problem of DeepFake detection. In addition to benchmark datasets, the model was further evaluated on AI-generated images from ChatGPT and Gemini, successfully detecting them as fake, demonstrating strong generalization.