Radhakrishnan Abilash, Sawant Tukaram Namdev, Raghuram Cheepurupalli, Railis Dani Jermisha, Singh Harjasdeep
Maria College of Engineering and Technology, Attoor, India.
Department of EXTC, Bharati Vidyapeeth College of Engineering, Navi Mumbai, India.
J Forensic Sci. 2025 Jul;70(4):1375-1391. doi: 10.1111/1556-4029.70069. Epub 2025 May 9.
In the digital age, accessible image manipulation raises concerns about authenticity, with forgery techniques threatening personal, journalistic, and security contexts. Detecting alterations is crucial for maintaining trust in visual content. A robust system capable of detecting various types of image forgeries, such as copy-move, splicing, and object removal, while minimizing false positives and negatives. Develop and implement robust feature extraction methods to identify key characteristics that differentiate forged images from authentic ones, focusing on both low-level and high-level features. The Two-dimensional maximum Shannon Entropy Median Filter (TSETMF) enhances image quality by reducing noise while preserving and enhancing details, which aids machine learning models in recognizing and identifying image forgeries. Multidimensional Spectral Hashing (MSH) enables efficient feature extraction by creating compact representations, thereby enhancing pattern recognition and boosting both speed and accuracy in detecting image forgeries within machine learning frameworks. Faster Region-Based Convolutional Neural Networks (FR-CNN) improve image forgery detection by swiftly identifying and localizing manipulated areas, enhancing feature extraction and accuracy for real-time forensic analysis. Machine learning approaches significantly enhance image forgery detection, with techniques like CNNs and MSH improving accuracy, processing speed, and robustness against diverse forgery methods, ensuring effective real-time analysis. The result shows that the proposed method significantly excelled, reaching an accuracy of 98.5%, alongside high precision (97.0%), recall (98.2%), and F1 score (98.1%), implemented using Python Colab. Future research can focus on developing more robust models, integrating unsupervised learning techniques, enhancing real-time detection capabilities, and exploring cross-domain applications to combat evolving image forgery methods effectively.
在数字时代，可便捷进行的图像操作引发了对图像真实性的担忧，伪造技术对个人、新闻和安全领域构成了威胁。检测图像篡改对于维持对视觉内容的信任至关重要。一个强大的系统应能够检测各种类型的图像伪造，如复制 - 粘贴、拼接和对象移除，同时尽量减少误报和漏报。开发并实施强大的特征提取方法，以识别区分伪造图像和真实图像的关键特征，重点关注低级和高级特征。二维最大香农熵中值滤波器（TSETMF）通过减少噪声来提高图像质量，同时保留并增强细节，这有助于机器学习模型识别和鉴定图像伪造。多维谱哈希（MSH）通过创建紧凑表示实现高效特征提取，从而增强模式识别能力，并提高机器学习框架内检测图像伪造的速度和准确性。基于区域的快速卷积神经网络（FR - CNN）通过快速识别和定位被篡改区域来改进图像伪造检测，增强特征提取能力并提高实时法医分析的准确性。机器学习方法显著增强了图像伪造检测能力，像卷积神经网络（CNNs）和多维谱哈希（MSH）等技术提高了准确性、处理速度以及针对各种伪造方法的鲁棒性，确保了有效的实时分析。结果表明，所提出的方法表现出色，准确率达到98.5%，同时具有高精度（97.0%）、召回率（98.2%）和F1分数（98.1%），该方法是使用Python Colab实现的。未来的研究可以专注于开发更强大的模型、整合无监督学习技术、增强实时检测能力以及探索跨领域应用，以有效应对不断演变的图像伪造方法。
