An Ocular Recognition System using Modified Facial Landmarks Extraction

• Main Authors: Szu-Han Tseng, 曾思翰
• Other Authors: Jing-Ming Guo
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/fffm9b

碩士 === 國立臺灣科技大學 === 電機工程系 === 103 === This thesis presents two strategies on ocular recognition system to achieve a high performance accuracy and low computational complexity. The first technique is the modified facial landmarks extraction which is the key aspect on feature extraction for identification system. The second strategy enables an innovative ocular recognition system. In terms of the facial landmarks extraction, the former random forest is improved with the proposed angle-split tactic to reduce the error rate. On the other hand, the facial landmarks extraction requires an additional process for the suppression of the noise interference. In this thesis, Gaussian blur filter is employed to alleviate the noise effect and to achieve a low error rate. In the ocular recognition system, the proposed method combines three various features, i.e., geometric, texture, and eye-fold texture, which can describe all bio-invariant properties of an ocular region. The support vector machine is then exploited to train a model to achieve a good recognition performance. As a result, the proposed system achieves a great flexibility in handling open eyes, blinking eyes, and closed eyes. Experimental results validate the successfulness and effectiveness of the proposed method over two standard image databases, i.e., LFPW and Helen databases. The proposed method reduces the error rate on the facial landmarks extraction stage. The performance of proposed method is also examined and investigated over another two image databases, i.e., CMU and Yale databases, which consist of open-and-blinking eyes scenarios. As documented in the experimental results, the proposed method offers a promising result in terms of recognition rate, and outperforms the former schemes. Thus, the proposed system can be regarded as an effective candidate in the biometric applications requiring real-time processing.