Face Restoration for Morphed Images Retouching

applied to morphed images, aiming to reduce or even remove any visible artifact and then to produce high-quality morphed images.However, this procedure is subjective, tedious, and time-consuming: these issues prevent the creation of large datasets with high-quality retouched morphed images that can be used to train or test Morphing Attack Detection (MAD) algorithms [4], i.e. automated tools explicitly devised to detect the presence of morphing in input images.More recently, morphing algorithms based on GANs [5] have been introduced as alternatives to traditional landmarkbased methods [6]- [8].The output of these GAN-based approaches typically does not present artifacts originated from issues with the facial landmark localization but still shows artifacts specific to the GAN-based generation procedure [9].Nonetheless, some approaches have limitations in generating high-resolution images, due to complexities in terms of memory requirements and training stability; moreover, controlling the final identity of the morphed face is more complex.Therefore, in this paper, we investigate the use of a Face Restoration [10] approach to automatically retouch morphed images created through a landmark-based morphing approach.Specifically, we verify the capability of CodeFormer [11] to retouch morphed images improving the visual quality while preserving identity.A fully automated retouch operation has the potential to eliminate the necessity for manual human intervention in handling morphed images and can simplify the process of generating extensive datasets comprising highquality images, which proves to be particularly beneficial for training and testing MAD algorithms.In addition, we believe this investigation can highlight the risk introduced by face restoration-based methods in creating morphed images that can easily deceive FRSs and humans. II. RELATED WORK A. Face MorphingFace Morphing is an image manipulation technique where one image undergoes a gradual transformation into another.This technique, applied in the scenario of manipulating electronic machine-readable travel documents (eMRTD) [1], allows the creation of facial images with a dual identity.Literature studies [2], [4] report that morphed images have the capability to circumvent both FRSs and human control, and then face morphing represents a severe security threat.Moreover, the proliferation of generative AI techniques for face Abstract-Creating a high-quality morphed image is a laborious and time-intensive endeavor due to the necessity of manual post-processing to eliminate typical artifacts produced by landmark-based morphing techniques.At the same time, morphed images of superior quality, without noticeable visual artifacts like ghosts, noise, or blurring, present heightened success probabilities to deceive human evaluators and commercial face verification s ystems.T herefore, i n t his p aper, w e investigate the use of Face Restoration to automatically retouch morphed images.Specifically, w e i nvestigate t he e fficacy of Co deFormer in removing artifacts and preserving the identity of the contributing subjects.An effective retouching method would allow the generation of large datasets containing high-quality retouched morphs, even starting from existing data, that are crucial for developing and evaluating the robustness of Morphing Attack Detection (MAD) algorithms.