Face recognition system
Download as PPT, PDF2 likes1,725 views
This document describes a face recognition system created by three students from Vietnam National University, Hanoi. The system uses a PDBNN algorithm for face detection and recognition. It first applies discrete Fourier transforms to input images before using neural network filters and merging to detect faces. Eye localization then precisely finds eyes before facial features are extracted using principal component analysis, Fisher's linear discriminant, or local feature analysis. The extracted feature vectors are input to face recognition to identify individuals. Potential applications of the system include security, access control, surveillance, photo tagging, and digital photography.
Face recognition system
- 1. VIETNAM NATIONAL UNIVERSITY, HANOI ĐẠI HỌC CÔNG NGHỆ University of Engineering & Technology Face Recognition System Members: Van-Ly Nguyen Van-Khai Ngo Ngoc-Sinh Nguyen
- 2. VIETNAM NATIONAL UNIVERSITY, HANOI ĐẠI HỌC CÔNG NGHỆ University of Engineering & Technology Face Recognition System Members: Van-Ly Nguyen Van-Khai Ngo Ngoc-Sinh Nguyen
- 3. ĐẠI HỌC CÔNG NGHỆ Outline Introduction Architecture Discrete Fourier Transform (DFT) Face Detection Eye Localization Facial Feature Extraction Face Recognition Applications Conclusion 5/23/12 Signals and Systems 3
- 4. ĐẠI HỌC CÔNG NGHỆ Introduction Are they the same people? 5/23/12 Signals and Systems 4
- 5. ĐẠI HỌC CÔNG NGHỆ Architecture Figure 1: System configuration of the PDBNN face recognition system 5/23/12 Signals and Systems 5
- 6. ĐẠI HỌC CÔNG NGHỆ Discrete Fourier Transform (DFT) Transformation of input images from time domain to frequency domain 5/23/12 Signals and Systems 6
- 7. ĐẠI HỌC CÔNG NGHỆ Discrete Fourier Transform (DFT) Fourier Transform in 2 – D images 5/23/12 Signals and Systems 7
- 8. ĐẠI HỌC CÔNG NGHỆ Discrete Fourier Transform (DFT) • Low frequencies contain much more information which is suitable for recognition than higher ones • Low frequencies are likely to be found at four corners of image spectrum 5/23/12 Signals and Systems 8
- 9. ĐẠI HỌC CÔNG NGHỆ Discrete Fourier Transform (DFT) Coefficient selection is one of the most important parameters in any recognition technique. Some coefficient selection methods: • Low frequency coefficient selection methods • Square selection methods • Curcular selection methods 5/23/12 Signals and Systems 9
- 10. ĐẠI HỌC CÔNG NGHỆ Face Detection Face Detection consists of two stages: • Neural – Network Filters • Merging overlapping detection Figure 2: The basic algorithm used for face detection. 5/23/12 Signals and Systems 10
- 11. ĐẠI HỌC CÔNG NGHỆ Face Detection The result of face detection: 5/23/12 Signals and Systems 11
- 12. ĐẠI HỌC CÔNG NGHỆ Eye Localization • Eye Localization is activated when face detection has found face in the input image. • Since the purpose of eye localization is to normalize facial patterns into a format the recognizer can accept, eye locations need to pinpoint with much higher precision than face location 5/23/12 Signals and Systems 12
- 13. ĐẠI HỌC CÔNG NGHỆ Facial Feature Extraction • Facial feature extraction techniques can be classified into two categories are feature – invariant approaches and template – based approaches Facial Feature Extraction Feature – Invariant Template-Based Approaches Approaches 5/23/12 Signals and Systems 13
- 14. ĐẠI HỌC CÔNG NGHỆ Facial Feature Extraction • Feature – Invariant Approaches This type of algorithm looks for structural features that exist even when the pose, viewpoint, or lighting condition vary The system can use various features including width of the head; distance from eyes to eyes, top of the head to eyes, eyes to the nose; and distance from eyes to the mouth • Template – based approaches The algorithm designs one or several standard face templates (usually frontal face template) either manually or by learning from examples in the image database 5/23/12 Signals and Systems 14
- 15. ĐẠI HỌC CÔNG NGHỆ Facial Feature Extraction One important issue for statistical template matching is the curse of dimensionality. Principal Component Analysis (PCA) Fisher's Linear Discriminant Efficient Extraction (PLD) Local Feature Analysis (LFA) 5/23/12 Signals and Systems 15
- 16. ĐẠI HỌC CÔNG NGHỆ Face Recognition Accept Feature Face Vectors Recognition Reject 5/23/12 Signals and Systems 16
- 17. ĐẠI HỌC CÔNG NGHỆ Applications Sercurity 5/23/12 Signals and Systems 17
- 18. ĐẠI HỌC CÔNG NGHỆ Applications Access control Surveillance 5/23/12 Signals and Systems 18
- 19. ĐẠI HỌC CÔNG NGHỆ Applications Photo tagging in Social Network Digital Photography 5/23/12 Signals and Systems 19
- 20. ĐẠI HỌC CÔNG NGHỆ Conclusion The System uses the PDBNN algorithm Face detection and Eye localization are two very important parts in the system PDBNN can perform these two processes at very high accuracy, more effective than other algorithms The system is more and more popular. 5/23/12 Signals and Systems 20
- 21. ĐẠI HỌC CÔNG NGHỆ Thank you & FAQs