A Dynamic Ensemble Selection of Deepfake Detectors Specialized for Individual Face Parts

The development of deepfake technology, based on deep learning, has made it easier to create images fake human faces that are indistinguishable from the real thing. Many methods and programs publicly available can be used maliciously, for example, by creating social media accounts with non-existent faces. To prevent misuse such images, several detection have been proposed as a countermeasure proven capable detecting deepfakes high accuracy when target model identified. However, existing approaches not robust partial editing and/or occlusion caused masks, glasses, or manual editing, all which lead an unacceptable drop in accuracy. In this paper, we propose novel approach dynamic configuration ensemble consists detectors. These detectors convolutional neural networks (CNNs) specialized detect focusing individual parts face. We demonstrate selection face selected CNN models is effective at realizing highly accurate even partly edited occluded images.