Research on Face Attribute Privacy Protection Method Based on Multi task Learning

With the widespread application of facial image recognition in fields such as finance, social media, and transportation, the privacy data of facial images is also frequently leaked. In order to effectively protect facial images, the protection of facial attributes is particularly important. Therefore, this article combines the method of multi task learning to divide deep learning models into low-level models and top-level models. The underlying model serves as a data feature extractor; The top-level model serves as a classifier, specifically responsible for three tasks in multi task learning: business classification, privacy protection, and image restoration. The goal of the business classification task is to identify user facial attributes; The goal of privacy protection tasks is to minimize the inclusion of privacy related feature information in data features as much as possible; The goal of image restoration task is to make it difficult to recover the original image containing privacy attributes from the extracted data features. In order to measure the privacy of the data feature extractor, a loss function is set for each task, and finally, the loss functions of these three tasks are fused to obtain a data feature extractor that can protect privacy. Finally, through experimental data analysis, this method ensures both the accuracy of facial image recognition and the leakage of facial attribute privacy data.