Design and Evaluation of a Novel Masked Face Feature Model for Face Recognition on Service Robot
Yunxi Zhang, Jiabao Wu, Ya Zhao, Zihan Wang, Jia Liu · 2024
Traditional face recognition techniques exhibit suboptimal accuracy when dealing with faces occluded by masks, particularly posing a challenge for masked face identification. Considering the application characteristics of the service robot, this paper proposes a face recognition method based on a special obscured face feature model. By leveraging the unique characteristics of masked faces, we construct a new feature model and curate a diverse dataset encompassing faces wearing both N95 and common medical masks, thereby enhancing the representativeness and variability of the dataset. Employing a cascade regression tree method for training and parameter tuning, we obtain an optimized regressor with minimal error. Face recognition is achieved through comparing feature vectors using Euclidean distance. Extensive testing on the curated dataset demonstrates the effectiveness of our proposed method, not only validating its superiority over traditional techniques but also showcasing its improved success rate in masked face recognition.