Camera Model Identification Based on Local Co-Occurrence Features

Abstract

We apply the algorithm proposed by Chen etal to identify camera models. The algorithm assumes that CFA pattern used by the device is GBRG. Local co-occurrence features are computed using multiple interpolation algorithms (example nearest neighbour, bilinear). A multi-class linear SVM is trained with these features and employed to classify the given image to one of the camera classes. Some observations have been made with respect to validation accuracy of the model and the results obtained on Kaggle.