Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{201256,
        author = {Peethani Satya Durga Rao and Y Dayanand Kumar},
        title = {A Scalable and Efficient Similarity Search Framework for Large-Scale Face Recognition Systems},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {3148-3152},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=201256},
        abstract = {These instructions give you guidelines for preparing papers for the International Journal of Innovative Research in Technology (IJIRT). Recent advances in deep learning have significantly improved the accuracy of face recognition systems through the use of facial embeddings; however, scalability remains a critical challenge in large-scale deployments. As the number of registered identities increases, searching within high-dimensional feature spaces becomes computationally expensive, reducing system efficiency. This study investigates the performance and scalability of multiple similarity search techniques for large-scale face recognition. Facial embeddings are generated using the FaceNet deep neural network, producing 128-dimensional feature vectors. Five search methods Linear Search, KD-Tree, Ball-Tree, FAISS Exact Search, and FAISS with Inverted File (IVF) indexing are evaluated using the Labeled Faces in the Wild (LFW) dataset, with database sizes expanded from 5,000 to 40,000 images. Experimental results show that traditional methods, including Linear Search and tree-based structures, experience significant performance degradation as dataset size increases due to computational complexity and high-dimensional data challenges. In contrast, FAISS-based methods, particularly IVF indexing, achieve substantially faster search times while maintaining strong recognition performance. The system achieves an Area Under the Curve (AUC) of approximately 0.974 and an Equal Error Rate (EER) of 0.07, demonstrating reliable discriminative capability. A limitation of this study is the use of dataset replication to simulate large-scale conditions, which may not fully capture real-world data diversity. Overall, the findings highlight the effectiveness of vector indexing techniques for building scalable and efficient face recognition systems.},
        keywords = {Face Recognition, Similarity Search, FAISS, Facial Embeddings, Biometric Systems},
        month = {May},
        }