A novel median filtering forensics based on principal component analysis network

As an important issue of forensic analysis, median filtering detection has drawn much attention in the decade. While several median filtering forensic methods have been proposed, they may face trouble when detecting median filtering on low-resolution or compressed images. In addition, the existing median filtering forensic methods mainly depend on the manually selected features, which makes these methods may not adapt to varieties of data. To solve these problems, convolution neural networks have been applied to learn features from the training database automatically. But the CNN-based method trains slowly and the parameters of it is hard to select. Thus, we proposed a PCANet-based method. And we test our trained model on several databases. The simulation shows that our proposed method achieves better performance, and trains much faster than CNN-based method.