On March 19th, Philipp K. G. Srock defended his Bachelor thesis on „Analysis and Detection of Facial Filters using Face Beauty“. Congratulations on the excellent work!
From the abstract:
Although facial attractiveness seems subjective, facial retouching is widely spread in social media, advertisements and even applied in professional photo studios to let the individuals appear younger, remove wrinkles and skin impurities and generally speaking: enhance the beauty. This is not a problem itself, but when retouched images are used as biometric samples and enrolled in a biometric system, it is one. Since previous work has proven facial retouching to be a challenge for face recognition systems, detection of facial retouching becomes increasingly necessary. This work analyzes changes in beauty assessment algorithms and quality assessment algorithms of retouched images, assesses different feature extraction methods for artificial intelligence based retouching detection and evaluates whether face beauty and face quality can be exploited to enhance the detection rate. An assumption can be made, that both provide useful features for the detection process. This study shows that even though face quality decreases while face beauty increases when a retouched sample is evaluated, face beauty and face quality do not measure the same features. Experiments conducted reveal that besides fusing multiple feature extraction methods, applying beauty scores improves the detection rate while face image quality has a very low impact when used to improve the detection rate. Detection results achieved on face filters not seen by the model during the training process are just above $1$\% average Equal-Error-Rate. Additionally it is shown that this rate can be further improved when using Machine Learning for the fusion process instead of a fusion function.
Goals:
The main goal of this work is to explore methods capable of improving digital image manipulation detection and more fine-grained research questions. This work focuses only on facial retouching detection. This is done by analyzing two beauty scoring mechanisms, face image quality and taking multiple feature extraction methods into account to fuse the scores generated by a deep learning classifier.