Design and Implementation of a Large-Scale High-Accuracy Multi-Face Recognition System Based on Vector Database and Super-Resolution Reconstruction

Facial recognition technology is now widely applied across various social contexts. However, in densely populated scenarios, traditional facial recognition algorithms struggle to accurately identify each individual due to the low resolution of each face. Concurrently, as the scale of facial databases increases, traditional facial recognition systems face difficulties in rapidly retrieving information from databases containing hundreds of thousands to millions of entries. To address these challenges, this paper constructs a large-scale and high-precision multi-face recognition system based on vector databases and super-resolution reconstruction technologies. Super-resolution reconstruction aims to enhance recognition accuracy by converting low-resolution images into high-resolution ones. Vector databases leverage parallel vector retrieval and GPU acceleration is to enable rapid searches within large-scale facial databases. The experiments results show that the single-face recognition achieves a Rank-1 accuracy of 93.9% and the multi-face recognition is reliable and accurate as well.