Recently, the research community spent a lot of effort to present face image
manipulation detection techniques, however, all the available schemes have their
own limitations and there is no global detection scheme. To avoid the problems
that can face the deep-learning based techniques, we suggest another direction of
research to implement a new Face Image Manipulation Detection (FIMD)
scheme which is based on face detection and image watermarking techniques.
The proposed FIMD scheme at the sender side has two stages where the first
stage is applied to detect and select the face area and the second stage is applied
to generate and embed the manipulation detection and localization data. The
proposed FIMD scheme at the receiver side has also two stages where the first
stage is applied to detect and select the face area and the second stage is applied
to extract the embedded data and check the authenticity of the received face
image. The experiments that have been conducted to check the performance of
the proposed FIMD scheme proved its efficiency in detecting different types of
face manipulations such as face swap, expression swap, attribute attacks, and
retouching attacks. The detection accuracy is 100 % and no false detection results
have been recorded, in addition, the scheme obtained promising results in terms
of visual quality of the watermarked face images and high embedding capacity.
The general comparison with the state-of-the-art detection schemes proved the
superiority of the proposed FIMD scheme.
أ د خميس عواد زيدان | Khamis A. Zidan | 10945