Morphing detection based on analysis of local spectrum
Type: Bachelor EE/CS
If you are interested please contact :
Face morphing is a technique to blend facial images of two or more subjects such that the result resembles both subjects. Face morphing attacks pose a serious risk for any face recognition system. Without automated morphing detection, state of the art face recognition systems are extremely vulnerable to morphing attacks. The most common approach to face morphing is by rst detecting landmarks in both contributing faces, then dene triangles in the images, determine an average geometry and map the averages of the textures in the triangles from the contributing faces to the averaged geometry.
Figure 1: From left to right: landmarks on face 1 and 2; triangulated geometry; morphed face.
Because two images are warped and blended together, we can expect that the local frequency content in the face images varies depending on local stretching or compression of the images. On the other hand, in the original images, we may expect that the local frequency content at least for the skin is relatively constant across the face. The aim of this assignment is to investigate if it is possible to detect these local variations in frequency content within a morphed face and investigate if they deviate suciently from the variation in non morphed images to allow for reliable detection of morphed images.