This study compares different face recognition techniques, evaluates them, and chooses the most effective technique based on a face recognition system that is comparatively complete. Find solutions to reduce storage needs while accelerating retrieval.

Face detection, face feature value extraction, and face feature value retrieval are the three stages of a full face retrieval system. Among them, face identification utilizing the MTCNN model is currently more used in the industry. For face feature value extraction and face feature value retrieval, there isn't, however, a unified processing approach available in the market right now. In order to determine a more effective strategy for extracting face feature values from FaceNet and InsightFace, we will examine and compare the face search accuracy in this work. To develop a more effective approach for retrieving face eigenvalues, this research will also examine and contrast the performance differences between using the Python loop and the vector database Milvus.

Finally, this study will employ MTCNN+InsightFace+Milvus to construct a comprehensive face retrieval system based on the aforementioned research findings of the aforementioned research. This article will also run benchmark tests on the Milvus database concurrently.