基于卷积神经网络的随机因子重采样图像检测

图像重采样检测是图像取证领域的重要任务，其目的是检测图像是否经过重采样操作。现有的基于深度学习的重采样检测方法大多只针对特定的重采样因子进行研究，而较少考虑重采样因子完全随机的情况。本文根据重采样操作中所涉及的插值技术原理设计了一组高效互补的图像预处理结构以避免图像内容的干扰，并通过可变形卷积层和高效通道注意力机制(efficient channel attention, ECA)分别提取和筛选重采样特征，从而有效提高了卷积神经网络整合提取不同重采样因子的重采样特征的能力。实验结果表明，无论对于未压缩的重采样图像还是JPEG压缩后处理的重采样图像，本文方法都可以有效检测，且预测准确率相比现有方法均有较大提升。

Image resampling detection with random factor based on convolutional neural network

Image resampling detection is an important task in the field of image forensic.The purpose is to detect whether the image is resampled.Current methods based on deep learning are mostly aimed at fixed resampling factors.However,they rarely consider the case that the resampling factors are completely random. In this paper,according to the principle of interpolation involved in resampling operation,an efficient preprocessing structure is designed to avoid the interference of image content.Then resampling features are extracted and screened by deformable convolutional layer and efficient channel attention mechanism respectively,so as to effectively improve the performance of convolutional neural network in extracting resampling features with different resampling factors.The experimental results show that whether for uncompressed resampling images or resampling images after JPEG compression, the method can detect effectively,and the prediction accuracy is greatly improved compared with the current methods.