Automated UNet++-Based Facial Wrinkle Segmentation System

In a variety of contexts, including skin diagnosis, plastic surgery consultations, and cosmetic advice, facial wrinkles are frequently used to assess the state of the skin or aging. Accurate wrinkle segmentation is necessary to detect wrinkled areas in order to handle face wrinkles in facial image analysis. Due to a lack of wrinkle data and an imbalance between wrinkle and non-wrinkle data, current deep learning-based techniques struggle to segregate small wrinkles. Thus, based on a UNet++ model, we present a novel approach to facial wrinkle segmentation in this study. In particular, we manually annotate fine wrinkles over the entire face to create a novel facial wrinkle dataset. Next, we use a facial landmark point extractor to extract just the skin area from the facial image. Finally, in order to mitigate the issue of class imbalance, we train the UNet++ model using both dice loss and focal loss. Using our dataset of facial wrinkles, we do extensive trials to verify the efficacy of the suggested approach. According to the experimental results, the suggested approach outperformed the most recent wrinkle segmentation method by 9.77% and 10.04% in terms of IoU and F1 score, respectively.