| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 103 === This thesis presents two techniques for a real-time fingerprint recognition system. The first technique delivers an efficient feature extraction with local invariant capability, namely Fast Binary Robust Local Feature (FBRLF), and the other technique presents the improved strategy for the fingerprint recognition system.
The FBRLF searches the stable features on an image which can be simply modeled using the adaptive Features from Accelerated Segment Test (FAST) corner detection with two user-defined parameters to yield stable features. To overcome the image noise, the Gaussian template is applied, and it is efficiently boosted by the integral image evaluation. In addition, the feature matching is conducted by incorporating the voting mechanism and lookup table method to achieve a high accuracy with low computational complexity. Experimental results demonstrate the superiority of the proposed method, making it suited for various applications such as pattern recognition, biometrics recognition systems, surveillance system, etc. In particular, the proposed method achieves a superior match rating in real-time fashion compared to that of the former competing schemes.
The traditional minutiae-based fingerprint recognition system requires a high computational complexity in the preprocessing stage. Most of them invlove unstable features, leading to a poor fingerprint recognition accuracy and heavy computational burden. The proposed fingerprint recognition system overcomes these issues by utilizing the proposed FBRLF. This system is built under Surface Pro 3, and is equipped with micro-optical fingerprint scanner to construct a real-time fingerprint recognition system. Two fingerprint databases are involved for the performance test of the proposed system. The first database is from manually generated fingerprint images, and the other is the FVC fingerprint standard database. Experimental results clearly demonstrate the superiority of the proposed method compared to the former schemes in terms of fingerprint recognition performance and processing efficiency.
|
|---|
