A Dempster-Shafer framework for decision fusion in image forensics

In this work a decision fusion strategy for image forensics is presented, based on Dempster-Shafer's Theory of Evidence. The goal is to automatically summarize the information provided by several image forensics tools, allowing both a binary and a soft interpretation of the global output produced. The proposed strategy is easily extendable to an arbitrary number of tools, it does not require that the output of the various tools be probabilistic and it takes into account available information about tools reliability. Comparison with logical disjunction- and SVM-based fusion shows an improvement in classification accuracy.