Facial Wrinkle Detection Based on DeepLabV3+ and Semi-Automatic Labelling Strategy

Facial skin wrinkles are positively correlated with physiological age and are an important feature of aging. The traditional wrinkle detection algorithms are influenced by facial features and image backgrounds, thus cannot distinguish hair, eyes and eyebrows well, and thus need to cut the facial region into multiple blocks. Motivated by the above challenges, this work presents a facial wrinkle detection algorithm based on DeepLabV3+ and a semi-automatic labelling strategy, which is featured by following procedures: (i) The algorithm first combines the facial texture features and rough annotation of wrinkles by dermatologists to generate the ground truth required for deep learning. (ii) A lightweight network, MobieNetV2, is employed as the backbone model to reduce the amount of network parameters and calculations. The constructed deep learning model is then trained using the original images and ground truth labels. (iii) The accuracy of various algorithms is evaluated using the Jaccard Similarity Index (JSI). The results demonstrate that the proposed method exhibits superior performance in wrinkle detection.