合成掩码技术生成的数据集
收藏资源简介:
本研究利用合成掩码技术,对现有的面部数据库如CelebA、LFW等进行改造,生成新的数据集用于掩码面部识别的研究。数据集包含多种面部图像,包括带掩码和不带掩码的图像,旨在解决传统面部识别系统在面对掩码面部时的识别问题。创建过程中,研究者使用了面部关键点预测和合成掩码形状生成技术,确保数据集的质量和适用性。该数据集的应用领域主要集中在提升面部识别系统在实际环境中的鲁棒性,特别是在全球疫情背景下,对于需要验证身份但面部被掩码遮挡的场景。
This study utilizes synthetic masking technology to modify existing facial databases including CelebA and LFW, thereby generating a new dataset for masked facial recognition research. The dataset encompasses diverse facial images, both with and without facial masks, aiming to resolve the recognition failures of traditional facial recognition systems when encountering masked faces. During the dataset development process, researchers adopted facial keypoint prediction and synthetic mask shape generation technologies to guarantee the dataset's quality and applicability. The main application of this dataset lies in improving the robustness of facial recognition systems in real-world environments, especially in scenarios requiring identity verification where faces are occluded by masks amid the global pandemic.

- 1Multi-Dataset Benchmarks for Masked Identification using Contrastive Representation Learning墨尔本大学 · 2021年



