The widespread adoption of face masks has generated significant challenges for traditional biometric systems. Conventional facial recognition methods, which depend largely on general facial features, exhibit a notable drop in performance when parts of the face are hidden. This issue is particularly important in the area of masked face recognition (1:N), where the system must effectively connect a masked probe image to a certain identity from a large collection of registered individuals. In contrast to verification (1:1 comparison), identification requires differentiating between numerous potential identities, significantly increasing the complexity of the task This paper proposes an approach to masked face identification that employs a VGG16 deep convolutional neural network as a pretrained feature extractor. The extracted embeddings are subsequently compared using cosine similarity within a Approximate Nearest Neighbor (ANN) search retrieval stage powered by Facebook AI Similarity Search (FAISS). By restructuring the identification process around robust embeddings and classical similarity measures, the method achieves competitive Rank-1 accuracy on the MFRD-80K dataset. The study demonstrates that even lightweight transfer learning strategies, when carefully optimized, can provide effective and computationally efficient solutions to the problem of masked face identification in real-world scenarios.