Camera Source Identification Using Discrete Cosine Transform Residue Features and Ensemble Classifier

Machine Learning based model building and classification has proved to be extremely effective for the camera source identification problem. In this paper, we have proposed a camera source identification methodology, based on extraction of the Discrete Cosine Transform Residual features, and subsequent Random Forest based ensemble classification with AdaBoost. We improve the classification accuracy by incorporating dimensionality reduction by Principal Component Analysis. Our experimental results on 10,507 images captured by ten cameras from the Dresden Image Database gives an average classification accuracy of 99.1%, and also show low overfitting trends when the constructed classifier is applied on a different image database.