Improving the Security of Image Manipulation Detection through One-and-a-half-class Multiple Classification.