Improved Morphed Face Images Datasets
Introduction
The Improved Morphed Face Images Dataset contains morphed face images based on the publicly available Face Research Lab London Set. These morphs are created using two different methods that address typical issues in automatic key-point-based face morphing pipelines, as described below.
One method is based on Neural Style Transfer and counters the image quality degeneration caused by the blending process in common face morphing pipelines. It can improve the quality, especially sharpness of the images, to be similar to the genuine input images.
Simple morph (left) vs. neural style transfer improved morphed face image (right): The improved image has a higher contrast and appears less dull than the simple morph. In particular, the freckles and the eye appear more sharp and realistic.
The other method improves the alignment process in face morphing pipelines. Most automatic face morphing pipelines align facial features based on only a limited number of key-points, which may also contain estimation errors. These errors and the limited number of key-points can lead to poorly aligned structures, causing noticeable artifacts known as ghosting artifacts in the final morph. Our pixel-wise alignment approach refines this initial warping, thereby preventing these artifacts.
Simple morph (left) vs. morph with pixel-wise alignment (right): The simple morph contains noticeable morphing artifacts in the left eye, in particular at the border of the iris, which is visible twice. In the improved version, the iris appears more natural, and even the specular highlights look more realistic.
All submitted papers or any publicly available text using the Improved Morphed Face Images Datasets must cite paper [2]
Related publication
[1] C. Seibold, A. Hilsmann, P. Eisert, Style Your Face Morph and Improve Your Face Morphing Attack Detector, 18th International Conference of the Biometrics Special Interest Group (BIOSIG), Darmstadt, Germany, Sep. 2019.
[2] Seibold C., Hilsmann A. and Eisert P., Towards Better Morphed Face Images without Ghosting Artifacts, Proc. VISAPP 2024, Rome, Feb. 2024.