Abstract
Facial recognition, as an efficient and convenient biometric technology, can be applied to automatic vending systems to achieve functions such as fast checkout and personalized recommendations. To improve the accuracy and processing speed of facial recognition technology, this study designs a facial recognition model for an automatic vending system with an improved principal component analysis method and the Holtling transform. This method reduces the dimensionality of facial features by introducing sample partitioning and histograms into principal component analysis to process facial data. On this basis, the Holtling transform is applied to process the reduced dimensional feature image to obtain the projection value of the face image, making the image easier to recognize. On the renderMe-360 and VoxCeleb2 datasets, the recognition accuracy of the model reached 96.32% and 98.24%, both higher than the comparison methods. The model had an average recognition accuracy of 94.388% in facial recognition from various angles, and showed significant efficiency advantages in feature face construction time and recognition time, especially in high noise conditions, demonstrating strong robustness. Therefore, the proposed model can improve the accuracy of facial recognition, as well as have faster processing speed and noise tolerance, providing new technical value for the intelligent development of automatic vending systems in the future.