Mobile Device based Intelligent Classification of Skin Severity Grades: Pigmentation, Wrinkles and Pores

The evaluation of the severity of skin conditions holds paramount significance in the realms of beauty, aesthetics, and dermatological diagnosis. This paper focuses on assessing the severity grades of various skin concerns, including pigmentation, wrinkles, or pores. Our proposed method is divided into 3 stages. The first stage is data collection and augmentation to increase the number of face data creating AIeS Skin E dataset by combining StyieGAN3 method and a set of hand-craft filtering rules. The next stage is face detection by YOLO5Face to focus on the skin analyzing area. For pigmentation and pores, information on the global full-face bounding box is analyzed while the wrinkles task focuses on analyzing 3 different local regions of the face including the forehead, eyes, and mouth. Finally, after passing through the classification model, one of three grades (good, average, or fair) is obtained for each skin condition of pigmentation, wrinkles, or pores. The accuracy of the method on AICS Skin E dataset increases compared to AICS Skin O dataset from 1.21% to 13.04% for pigmentation, wrinkles, and pores. The classification model's accuracy on mobile devices decreases slightly for each skin condition and the speed is from 5.5 to 6.1 fps.