1997 Fiscal Year Final Research Report Summary
Designing Security Systems Using Face Recognition Technique
Project/Area Number |
08458086
|
Research Category |
Grant-in-Aid for Scientific Research (B)
|
Allocation Type | Single-year Grants |
Section | 一般 |
Research Field |
情報システム学(含情報図書館学)
|
Research Institution | CHIBA UNIVERSITY |
Principal Investigator |
YAHAGI Takashi Chiba University, Faculty of Engineering, Professor, 工学部, 教授 (90009530)
|
Co-Investigator(Kenkyū-buntansha) |
YAMAMOTO Kazuo Chiba University, Faculty of Engineering, Researcher, 工学部, 教務職員 (30110290)
RO Kenmei Chiba University, Graduate School of Sience and Technology, Assistant, 大学院・自然科学研究科, 助手 (80251180)
|
Project Period (FY) |
1996 – 1997
|
Keywords | Security System / Parallel System / Fuzzy Theory / Face Recognition / Recognition Rate / Small Scall Neural Network |
Research Abstract |
Recently, there are different levels of security which are becoming the main problem in the field of information systems. When face recognition is employed as the key in security problem good results can be obtained using neural networks only for a few number of registered individuals. However, with the increase in the registered individuals recognition rate falls and the training time increases because of the huge amount of facial information. In this work, we introduce small scale parallel neurals networks governed by Fuzzy theory to get rid of the aforementioned problem. In the proposed method, the input image is first classified into some categories based on the pattern matching theory. Then, using the Fuzzy theory the closest patterns are attributed into some categories. In the second step, the final decision pertaining to the images belonging to each category is reached based on the small scale neural networks. When a new category is added to the sytem it does not affect the existing neural networks rather a neural network that includes a new category is created to append to the pre-existing system. Doing this there is no need of retraing the whole network. The small scale neural network has 256 neurons in the input layr, 128 neurons in the hidden layr and, 6 neurons in the output layr. The backpropagation algorithm is used to train this network. The test results (for 25 registered individuals) demonstrate that a recognition rate of 99.33% can be obtained using the proposed method.
|