A Systematic Review of Face Recognition Attendance Systems: Anti-Spoofing Integration, Intelligent Automation, and Research Gaps

A Systematic Review of Face Recognition Attendance Systems: Anti-Spoofing Integration, Intelligent Automation, and Research Gaps

Yash R. Pounikar, Ayushi A. Bhadade, Amit R. Khotele, Yash S. Bhoyar, Varun D. Khadse, Bharat S. Dhak

Department of Artificial Intelligence Priyadarshini Bhagwati College of Engineering, Nagpur

R.T.M Nagpur University, Nagpur, Maharashtra, India

Abstract – This paper presents a comprehensive literature review of existing face recognition-based attendance systems and anti- spoofing techniques to identify research gaps and propose solutions. We analyze ten significant research works covering face recognition algorithms, liveness detection methods, system architectures, and deployment frameworks. Our review reveals that while individual components have been well-researched, existing systems lack comprehensive integration of high-accuracy recognition with multi- layered anti-spoofing, real-time analytics, and intelligent attendance logic in a unified framework. Most implementations focus on either recognition accuracy or liveness detection but rarely combine both effectively. We identify critical gaps including absence of intelligent check-in and check-out logic, limited real-time analytics capabilities, inadequate multi-modal liveness detection, and lack of scalable web-based architectures. Based on these findings, we propose an integrated system combining ArcFace recognition, four- layer liveness detection, Flask-based real-time streaming, PostgreSQL with intelligent attendance algorithms, and Node.js powered analytics dashboard to address identified limitations.

Keywords face recognition, attendance system, literature review, anti-spoofing, liveness detection, ArcFace, research gaps, proposed solution

1. INTRODUCTION

   Attendance management remains a fundamental administrative challenge in educational institutions and corporate environments. Traditional methods including manual roll calls, RFID cards, and fingerprint scanners suffer from time consumption, proxy attendance, hygiene concerns, and maintenance overhead. Face recognition technology offers a contactless alternative, but practical deployment requires addressing multiple technical challenges including recognition accuracy, spoofing prevention, real-time processing, and intelligent attendance logic.

   This literature review examines ten significant research contributions in face recognition-based attendance systems to identify existing capabilities and research gaps. Our objective is to understand what has been achieved, what limitations persist, and how a comprehensive system can address these gaps. We analyze each work systematically, examining their contributions and limitations, then propose an integrated solution that overcomes identified shortcomings.

   The review covers face recognition algorithms, particularly ArcFace architecture, various liveness detection approaches, system implementation frameworks, and database management techniques. By synthesizing findings across these works, we identify opportunities for creating a more robust, secure, and user-friendly attendance system suitable for real-world deployment in educational institutions.

2. LITERATURE REVIEW

   1. ArcFace: Additive Angular Margin Loss for Deep Face Recognition

      Deng et al. introduced ArcFace as an additive angular margin loss function for deep face recognition. The method achieved state-of-the-art performance with 99.83 percent verification accuracy on the Labeled Faces in the Wild dataset. ArcFace enhances feature discrimination by adding angular margin penalties in softmax loss, resulting in compact intra-class distributions and well-separated inter-class boundaries. The model generates 512-dimensional embeddings that effectively capture facial characteristics while remaining robust to lighting, pose, and expression variations.

      While ArcFace provides excellent recognition accuracy, the paper focuses purely on the recognition algorithm without addressing practical deployment considerations. It does not discuss liveness detection, real-time video processing pipelines, attendance logic implementation, or database integration. The research validates performance on static benchmark datasets but lacks guidance for building complete attendance systems. Additionally, no discussion of handling continuous video streams, preventing duplicate entries, or managing check-in and check-out workflows appears in the work.

   2. Face Recognition-Based Mass Attendance Using YOLOv5 and ArcFace

      Faruque et al. combined YOLOv5 for face detection with ArcFace for recognition in mass attendance scenarios. Their two-stage pipeline first detects multiple faces simultaneously

      using YOLOv5, then applies ArcFace for identification. The system achieved 95.2 percent accuracy in crowded environments, demonstrating effectiveness for educational settings where students enter classrooms in groups.

      The system focuses on detection and recognition but provides limited information about anti-spoofing mechanisms. No comprehensive liveness detection appears to be implemented, leaving the system vulnerable to presentation attacks using photographs or videos. The paper does not address intelligent attendance algorithms that distinguish between check-in and check-out, prevent duplicate entries, or validate minimum stay duration. Database schema design and real-time analytics capabilities are not discussed. The system appears to mark attendance without temporal logic or fraud prevention mechanisms.

   3. Face Liveness Detection Using Artificial Intelligence Techniques

      Hasan et al. provided a systematic literature review of liveness detection methods categorized into texture-based, motion-based, and hybrid approaches. They discussed various techniques including texture analysis examining local patterns, motion-based tracking of eye blinks and micro- expressions, and hybrid systems combining multiple modalities. Their analysis indicated that multi-modal approaches achieve detection rates exceeding 98 percent.

      While comprehensive in reviewing liveness detection methods, this work is a survey paper that does not present an implemented system. It lacks integration with actual face recognition pipelines, real-time video processing frameworks, or database systems. The paper does not address how to effectively combine multiple liveness checks in a single system, what thresholds work best for different checks, or how to balance security with user experience. No discussion of implementing these techniques in web-based attendance systems or integrating with recognition algorithms appears in the review.

   4. Real-Time Face Liveness Detection and Face Anti- spoofing Using Deep Learning

      Zawar and Chakkarwar proposed real-time liveness detection combining convolutional neural networks with Eye Aspect Ratio calculations for blink detection. Their hybrid approach integrates deep learning-based texture analysis with traditional geometric features, achieving 96.8 percent accuracy while maintaining real-time performance.

      The system implements only two liveness checksCNN- based texture analysis and EAR-based blink detection. This dual-modal approach may be insufficient against sophisticated presentation attacks. Other detection methods such as color diversity analysis in HSV space or moiré pattern detection using FFT are not included. The paper does not discuss integration with complete attendance systems including database management, attendance loic, or analytics dashboards. No information about handling check- in and check-out workflows or preventing duplicate entries within time windows appears in the work.

   5. A Robust Anti-Spoofing Technique for Face Liveness Detection with Morphological Operations

      Singh and Arora developed anti-spoofing techniques using morphological operations to analyze structural differences between live and fake faces. Their method examines texture patterns and edge characteristics through morphological opening and closing operations. Testing under varying lighting conditions validated reliability for practical deployment.

      The technique focuses solely on morphological operations for liveness detection without discussing other complementary methods. Relying on a single detection approach limits robustness against diverse attack types. The paper does not address integration with face recognition algorithms, real-time video streaming, or attendance management logic. No discussion of system architecture, database design, or web-based interface implementation appears. The work validates the anti-spoofing technique in isolation but does not demonstrate integration into a complete attendance system.

   6. Face Recognition Smart Attendance System using Deep Transfer Learning

      Alhanaee et al. presented a smart attendance system leveraging deep transfer learning. Pre-trained CNN models were fine-tuned on institution-specific datasets, reducing training time and computational requirements. The system achieved high recognition accuracy in classroom environments.

      While demonstrating effective use of transfer learning for recognition, the paper provides minimal detail about anti- spoofing mechanisms. Liveness detection appears either absent or inadequately described. The system does not discuss intelligent attendance algorithms that handle check-in and check-out logic, time-based validation, or duplicate prevention. No information about real-time analytics, dashboard visualization, or administrative tools appears. The database schema and attendance record management strategies are not detailed. The focus remains primarily on improving recognition accuracy through transfer learning without addressing the broader system requirements.

   7. Student Attendance System Using Face Recognition

      This IEEE conference paper describes a face recognition- based attendance system designed to automate student attendance. The system captures facial images, performs recognition, and records attendance in real time. Results highlight improved accuracy, reduced manual effort, and prevention of proxy attendance.

      The paper provides a high-level overview without sufficient technical depth. Anti-spoofing mechanisms are not discussed, leaving the system potentially vulnerable to presentation attacks. No details about the attendance algorithm logic, such as how check-ins and check-outs are managed, minimum duration validation, or cooldown periods appear. The database structure and query optimization

      strategies are not explained. Real-time analytics capabilities including trend analysis, hourly distribution patterns, or branch-wise comparisons are absent. The system appears to be a basic implementation without the sophisticated features required for enterprise deployment.

   8. LISTEN/NOTIFY for Real-Time Applications

      PostgreSQL documentation explains the LISTEN and NOTIFY mechanism for building real-time, event-driven applications. This feature enables efficient communication between database events and application layers without constant polling, making it effective for real-time notification systems.

      While PostgreSQL provides robust database capabilities and real-time notification features, the documentation focuses on database functionality without addressing attendance system implementation. No discussion of face recognition integration, liveness detection, or attendance logic appears. The documentation does not provide complete system architecture examples showing how to combine LISTEN and NOTIFY with video streaming, face processing pipelines, and web interfaces. Developers must bridge the gap between database capabilities and complete attendance system requirements independently.

   9. InsightFace: 2D and 3D Face Analysis Project

      InsightFace is an open-source framework providing state- of-the-art face recognition models including ArcFace. It supports high-performance facial feature extraction for large- scale applications and is widely used in academic and industrial systems.

      InsightFace provides excellent face recognition capabilities but is a library rather than a complete system. It does not include anti-spoofing mechanisms, attendance management logic, database integration, or web interfaces. Developers must build these components independently. No guidance for implementing intelligent attendance algorithms, real-time analytics dashboards, or preventing duplicate entries appears in the framework. While InsightFace handles recognition effectively, creating a production-ready attendance system requires substantial additional development work.

   10. Face Recognition with OpenCV

   OpenCV documentation provides comprehensive tools for face detection and recognition using classical and deep learning-based approaches. It supports real-time image processing and integrates with Python and C++, making it a foundational framework for computer vision applications.

   OpenCV provides low-level computer vision tools but does not offer high-level attendance system functionality. Developers must implement liveness detection, attendance logic, database management, and web interfaces from scratch. No built-in support for intelligent check-in and check-out workflows, duplicate prevention, or analytics dashboards exist. While OpenCV handles video capture and

   image processing effectively, building a complete attendance system requires integrating multiple additional technologies and implementing substantial custom logic.

3. IDENTIFIED RESEARCH GAPS

   Based on the comprehensive review of existing literature, we identify the following critical research gaps:

   1. Lack of Integrated Multi-Modal Liveness Detection

      Most systems implement single or dual-modal liveness detection, which proves insufficient against sophisticated presentation attacks. No reviewed work integrates all four complementary methods: Eye Aspect Ratio blink detection, texture analysis using Laplacian variance, HSV-based color diversity analysis, and FFT-based moiré pattern detection in a unified framework with weighted confidence scoring.

   2. Absence of Intelligent Attendance Logic

      Existing systems lack sophisticated attendance algorithms that automatically distinguish between check-in and check- out events. Key missing features include automatic check-in upon first daily recognition, automatic check-out after minimum duration validation, duplicate prevention within configurable time windows, and cooldown periods preventing immediate re-entry after exit.

   3. Limited Real-Time Analytics Capabilities

      Most implementations focus on basic attendance recording without comprehensive analytics. Missing capabilities include real-time dashboard with live statistics, temporal trend analysis across multiple time periods, branch- wise and category-wise comparisons, hourly attendance distribution patterns, and WebSocket-based instant notifications.

   4. Incomplete System Integration

      Research works typically focus on individual components rather than end-to-end integration. Complete systems require seamless integration of face recognition with liveness detection, video streaming with database updates, attendance logic with analytics dashboards, and backend processing with responsive web interfaces. No reviewed work demonstrates this comprehensive integration.

   5. Limited Scalability Considerations

   Most systems do not address scalability requirements including efficient database indexing strategies, connection pooling for concurrent users, frame skipping optimization for processing efficiency, and distributed architecture supporting multiple entry points.

4. PROPOSED SOLUTION

   To address identified research gaps, we propose an intelligent face recognition-based attendance management system with comprehensive anti-spoofing capabilities. The proposed system integrates state-of-the-art technologies into

   a unified framework addressing all limitations identified in the literature review.

   1. High-Accuracy Face Recognition

      The proposed system will utilize InsightFaces ArcFace model to generate 512-dimensional facial embeddings with cosine similarity matching. This approach is expected to achieve recognition accuracy exceeding 95 percent while maintaining sub-second response times. Unlike previous works that use ArcFace in isolation, our system will integrate it within a complete real-time video processing pipeline with optimized frame handling.

   2. Comprehensive Four-Layer Liveness Detection

      Addressing the gap in multi-modal anti-spoofing, the proposed system will implement four complementary liveness checks:

      Eye Aspect Ratio Analysis: Using dlibs 68-point facial landmark predictor to detect natural blinks with configurable EAR thresholds.

      Texture Analysis: Calculating Laplacian variance on grayscale face images to detect printed photographs with threshold optimization.

      Color Diversity Analysis: Analyzing standard deviation in HSV color channels to identify screen-based attacks with sensitivity tuning.

      Moiré Pattern Detection: Applying Fast Fourier Transform to detect periodic patterns characteristic of photographed screens.

      Each check will generate a confidence score, and weighted combination will determine final liveness verdict. This multi- layered approach is expected to achieve anti-spoofing detection rates exceeding 98 percent, significantly surpassing systems using single or dual-modal detection.

   3. Intelligent Attendance Algorithm

      The proposed system will implement sophisticated attendance logic addressing gaps in existing systems:

      Automatic Check-In: Upon recognizing a face, the system will query the database for existing entries on the current date. If none exists, it will automatically create a check-in record with timestamp.

      Automatic Check-Out: If an entry exists with present status, the system will validate that minimum stay duration has been met (configurable, default 2 minutes). Upon validation, it will record check-out time and calculate total duration automatically.

      Duplicate Prevention: The system will enforce a cooldown period (default 20 seconds) after exit before allowing next entry, preventing multiple spurious detections from continuous video frames.

      Time Validation: All operations will include timestamp validation ensuring logical consistency and preventing backdated entries.

   4. Real-Time Video Streaming Architecture

      The proposed system will use Flask to serve real-time video streams using MJPEG format with multipart responses. The architecture will process frames efficiently through frame skipping (every third frame) while maintaining smooth visual feedback. Face detection, recognition, and liveness checks will be executed in the processing pipeline with results overlayed on the video stream. This addresses gaps in previous works that either lacked real-time capabilities or did not explain streaming implementation.

   5. Comprehensive Analytics Dashboard

      The proposed system will include a Node.js and Express.js powered analytics dashboard providing:

      Real-Time Statistics: Total visitors, currently present count, and category-wise breakdowns updated live.

      Trend Analysis: Weekly, monthly, and yearly attendance patterns with interactive Chart.js visualizations.

      Hourly Distribution: Time-based attendance patterns showing peak and off-peak periods.

      Branch-Wise Comparison: Departmental attendance comparisons for institutional insights.

      WebSocket Integration: Socket.io will provide instant notifications when attendance events occur.

      PostgreSQL LISTEN/NOTIFY: Database-level event propagation will ensure all connected clients receive immediate updates.

      This comprehensive analytics capability addresses the gap identified in existing systems that focus on recording attendance without providing actionable insights.

   6. Scalable Database Architecture

      The proposed system will use PostgreSQL with optimized schema design including:

      Indexed Columns: Indexes on erpid, date, and status will ensure fast queries even with large datasets.

      Generated Columns: Automatic duration calculation using time_out minus time_in without application logic.

      Unique Constraints: Database-level enforcement will prevent duplicate entries.

      Connection Pooling: Efficient connection management will support concurrent users without resource exhaustion.

   7. Secure and User-Friendly Interface

   The proposed system will provide a responsive web interface built with HTML5, CSS3, and JavaScript. The frontend will display live video with recognition overlays, show recognized person details including ERP ID and department, provide instant notifications for entry and exit events, and offer searchable and filterable attendance logs. Context managers will ensure proper resource cleanup and error handling will maintain system stability.

5. COMPARATIVE ANALYSIS

   Feature	Previous Works	Proposed System
   Multi-Modal Liveness	Single or dual modal	Four-layer detection
   Check-In/Out Logic	Manual or absent	Fully automatic
   Real-Time Analytics	Limited or absent	Comprehensive dashboard
   Video Streaming	Basic or undocumented	Flask MJPEG streaming
   Database Integration	Basic recording	PostgreSQL with LISTEN/NOTIFY
   Duplicate Prevention	Not addressed	Time-based validation
   Recognition Accuracy	95-96%	>95% with optimization
   Anti-Spoofing Rate	96-98%	>98% multi-modal
   Scalability Design	Not addressed	Indexed, pooled architecture

   The table below compares the proposed systems capabilities against limitations identified in reviewed literature:

   User Satisfaction: Contactless, fast, and reliable attendance marking improving user experience compared to traditional methods.

   VII. CONCLUSION

   This literature review analyzed ten significant research contributions in face recognition-based attendance systems and anti-spoofing techniques. While individual components have been well-researched, we identified critical gaps including lack of integrated multi-modal liveness detection, absence of intelligent attendance logic, limited real-time analytics, incomplete system integration, and insufficient scalability considerations.

   To address these gaps, we propose a comprehensive system integrating ArcFace recognition, four-layer liveness detection, intelligent attendance algorithms, Flask-based video streaming, PostgreSQL with real-time notifications, and Node.j powered analytics dashboard. This integrated approach overcomes limitations identified in existing literature, providing a production-ready solution suitable for educational institutions and corporate environments.

   Our proposed system represents a significant advancement by combining best practices from reviewed literature while addressing their individual limitations through comprehensive integration. Future work will be implemented, test, and validate the proposed system in real- world educational settings, providing empirical evidence for our design decisions and identifying opportunities for further improvement.

6. EXPECTED OUTCOMES

Based on our proposed architecture, we anticipate the system will deliver:

High Recognition Accuracy: Greater than 95 percent recognition accuracy under real-world conditions including varied lighting and angles.

Robust Anti-Spoofing: Detection rate exceeding 98 percent against common presentation attacks including photographs, videos, and screen displays.

Real-Time Performance: Sub-second processing time from face detection through liveness verification to database update.

Intelligent Automation: Zero manual intervention for check- in and check-out with automatic duration calculation and status management.

Comprehensive Analytics: Real-time insights with multiple visualization types supporting data-driven decision making.

Scalability: Support for hundreds of concurrent users with response time degradation under 10 percent.

ACKNOWLEDGMENT

We thank Prof. B.S. Dhak for guidance in conducting this literature review. We acknowledge the Department of Artificial Intelligence at Priyadarshini Bhagwati College of Engineering for providing research resources. We also thank Dr. A.M. Shende, Principal of PBCOE, for institutional support.

REFERENCES

1. J. Deng, J. Guo, N. Xue, and S. Zafeiriou, ArcFace: Additive Angular Margin Loss for Deep Face Recognition, arXiv:1801.07698, 2019.

2. O. Faruque, F.H. Siddiqui, and S.B. Noor, Face Recognition-Based Mass Attendance Using YOLOv5 and ArcFace, Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, vol. 490, Springer, 2023.

3. M.K. Hasan et al., Face Liveness Detection Using Artificial Intelligence Techniques: A Systematic Literature Review, MDPI Sensors, vol. 7, no. 1, pp. 37, 2023.

4. R. Zawar and V. Chakkarwar, Real-Time Face Liveness Detection and Face Anti-spoofing Using Deep Learning, Proceedings of ACVAIT 2022, Atlantis Press, 2023.

5. M. Singh and A.S. Arora, A Robust Anti-Spoofing Technique for Face Liveness Detection with Morphological Operations, Optik, vol. 139, Elsevier, 2017.

6. K. Alhanaee et al., Face Recognition Smart Attendance System using Deep Transfer Learning, Procedia Computer Science, vol. 190, 2021.

7. IEEE, Student Attendance System using Face Recognition, IEEE

   Conference Publication, 2020.

8. PostgreSQL Documentation, LISTEN/NOTIFY for Real-Time Applications, Crunchy Data, 2023.

9. InsightFace, InsightFace: 2D and 3D Face Analysis Project, GitHub Repository, 2023.

10. OpenCV, Face Recognition with OpenCV, OpenCV Documentation, 2023.