Objective To develop a mask suitability correlation model based on facial features extracted from facial images using facial recognition technology, enabling rapid mask recommendations for users and reducing infection risk.

Methods Frontal facial photographs of volunteers were collected, and mask sealing performance tests were conducted to obtain facial feature dimensions and mask suitability data. Dimensionality reduction and clustering were applied to the facial feature data, and a correlation model was established between facial clusters and mask sealing performance data. This model facilitates rapid mask recommendations based on facial images.

Results Facial image data from 245 volunteers were collected. Each volunteer randomly selected 2–3 masks from five common types for sealing performance tests, resulting in a total of 589 mask sealing performance data points, with each mask type tested 115–125 times. Statistical analysis revealed that two masks, ENVIP3DF (84.3%) and 801-N95 (68.9%), had higher passing rates, with a statistically significant difference (χ²=127.546, P<0.05). The majority of the data was used for model training, and a small portion was used for validation. Validation results showed that for volunteer 1, the test scores and recommendation degrees for two randomly selected masks were ENVIP3DF (200, 95.0%) and 801-N95 (145, 94.6%), respectively; for volunteer 5, the scores and recommendation degrees were 801-N95 (152, 64.7%) and 9132 (52, 39.4%), respectively. For each volunteer, there was a significant positive correlation between the actual mask sealing performance test score and the proposed mask suitability recommendation correlation model (r=0.642, P<0.05). The proposed method can accurately predict mask sealing test results based on facial image processing data.

Conclusions This method can rapidly recommend optimal masks based on facial images and is suitable for emergencies or situations where standard mask sealing performance testing is inconvenient.