RFID ACCESS AUTHORIZATION BY FACE RECOGNITION 報告學生：翁偉傑 1 Proceedings of the Eighth International Conference on Machine Learning and Cybernetics, Baoding,

RFID ACCESS AUTHORIZATION BY FACE

RECOGNITION

報告學生：翁偉傑

1 Proceedings of the Eighth International Conference on Machine Learning and Cybernetics, Baoding, 12-15 July 2009

OUTLINE

Introduction

Security System for RFID Access Control

Face Feature Extraction with SIFT

L-GEM Trained RBFNN based Face Recognizer

Experimental Results

Conclusions 2

INTRODUCTION

Radio Frequency Identification (RFID)的主要缺點，任何人都可以得到該卡存取。

本研究提出了一種基於神經網絡的人臉識別系統，

本研究提出了一個 Localized Generalization Error Model (L-GEM)的徑向 Radial Basis Function Neural Network (RBFNN) 人臉辨識系統，以提高安全性的 RFID卡的系統。 3

SECURITY SYSTEM FOR RFID ACCESS CONTROL (1/2)

4

The Processes of Security System Training

SECURITY SYSTEM FOR RFID ACCESS CONTROL (2/2)

5

The Processes of the Security System

FACE FEATURE EXTRACTION WITH SIFT (1/3)

6

範例：

SIFT 它的穩定性和精度高優於其他描述。

FACE FEATURE EXTRACTION WITH SIFT (2/3)

7

Local Feature Descriptors of Two Images of the Same Person

FACE FEATURE EXTRACTION WITH SIFT (3/3)

8The Local Feature Descriptors of Two Images of Two Different People

L-GEM TRAINED RBFNN BASED FACE RECOGNIZER (1/3)

9

The RBFNN is trained as follows:

1. Store the RFID card owner information in the database and take 10 images of the owner

2. L-GEM trained RBFNN to recognize face of person.

3. The trained RBFNN is then stored in the database and associated with the card owner.

L-GEM TRAINED RBFNN BASED FACE RECOGNIZER (2/3)

10is a general framework to estimate the localized generalization error of a classifier

where N, Remp and SM denote the number of training samples, the training mean square error and the stochastic sensitivity measure of RBFNN, respectively.

L-GEM TRAINED RBFNN BASED FACE RECOGNIZER (3/3)

11

The procedures for the face recognition after reading the RFID card ID is as follows:

1. Fetch the card owner’s RBFNN from the database

2. Face Detection by Adaboost and output a small image with face only

3. Extract local feature descriptor from the detected face image

4. Classify the face by the RBFNN trained by the L-GEM

5. If the face owner does not match the RFID card owner, alert security, otherwise end

EXPERIMENTAL RESULTS (1/4)

In this experiment, we assume that there are 3 users of the security system. Each of them holds an RFID card. The system is built to verify whether the card holder is the card owner. We name these 3 people as Person A, Person B and Person C for convenience.

以 10張為訓練圖像。

實驗 90次，分為 30 次 ( 個人 )和 60 次 ( 混合 ) 。12

EXPERIMENTAL RESULTS (2/4)

13

Training Images of Person A

EXPERIMENTAL RESULTS (3/4)

14

Moving in the entrance A face is detected

Tracing the detected face The Face leaving

EXPERIMENTAL RESULTS (4/4)

15

Testing Results of the Security System

CONCLUSIONS

This research proposed a security system combining RFID card access control with face recognition by RBFNN trained by the L-GEM.

enhance security of RFID card based access control systems

Can only recognize one person per access.

16