Facial image de-identification using active appearance model
Jiří Přinosil, Petr Kříž, Kamil Říha, Malay Kishore Dutta, Ashish Issac · 2017
Nowadays, the task of preservation of personal data present in public audio-visual records becomes more important. This paper describes an image processing method for de-identification of visual personal data based on a facial image. There are many primitive methods with high robustness but image quality of de-identified data is usually very poor. The advantages of this proposed method include especially very good results in quality preservation of the original image data and the possibility of a partial image reconstruction from de-identified data. The main objective has been achieved by using Active Appearance Model algorithm tested on a created database with the overall success rate of automatic recognition tool around 0.8 %. The de-identified image data look also authentic and it is possible to determine the expression or gender of the de-identified person.