Cristian Botezatu defended his Master Thesis on Impact of Selfie Filters on Face Recognition
Each individual has distinctive physiological and behavioural characteristics, making it possible for biometric systems to use this information for recognition purposes. Among other biometric systems, face recognition is widely accepted, convenient and accurate, seeing significant performance enhancement since the appearance of deep learning. Despite the nearly perfect recognition performance of state-of-the-art face recognition systems, there is still concern with regards to their reliability when being exposed to occluded faces.
Inline with the aforementioned information, the goal of the thesis is to assess the effect of selfie filters on face recognition systems, and to provide an algorithm for reconstructing occluded facial parts with the aim of improving face recognition performance. To be able to engage in such an investigation, an appropriate dataset of facial images is required. To this end, popular mobile applications are used to create selfie filtered images. The created datasets of selfie filtered facial images, are used to evaluate the impact of selfie filters on face recognition systems, by comparing the performance of state-of-the-art face recognition systems on unaltered and corresponding selfie filtered facial images.
The results show that selfie filters affect both the tested commercial and open-source systems, especially on facial images where the eye or nose region is occluded. Furthermore, selfie filters of high facial coverage have shown to be the most challenging, significantly declining face detection, face quality assessment and face recognition performance.