基於卷積神經網路方法建構虹膜辨識系統

Constructing the Iris Recognition System based on Convolutional Neural Network Approach

黃千芳(Chien-Fang Huang)

指導教授 ： 陳牧言

國立臺中科技大學/資訊與流通學院/資訊管理系碩士班/碩士(2020年)

摘要

台灣是個熱衷於賽鴿的國家，龐大的獎金及檯面下的賭注成為極大誘因，誘使不少不肖之徒企圖利用作弊、竊鴿等方式非法取得利益，導致賽事公平性及法律問題層出不窮。其中賽鴿的身分驗證問題一直以來為人詬病，早期賽鴿協會仰賴傳統腳環及人工辨別來驗證賽鴿身分，缺乏效率及公信度，後加入無線射頻辨識(radio frequency identification，RFID)技術改良為電子腳環，但此加密方法仍有被破解的安全疑慮，近年來又有賽鴿DNA檢定的產業興起，雖能達到極高的準確度但成本昂貴且無法即時比對。因此本論文目的在於提出一個成本相對較低、驗證速度更快，且準確度具有一定水準的賽鴿身分辨識系統。基於鴿子虹膜與人類虹膜同樣具有高度獨特性且穩定不會改變的假設，本研究認為使用鴿子虹膜進行生物辨識是一種能夠準確驗證賽鴿身份的方法，但要取得鴿子虹膜圖像相比人類困難許多，因此目前少有鴿子虹膜識別的相關研究。本論文的貢獻為透過改良的YOLO物體檢測演算法自動檢測鴿子虹膜區域，取代傳統虹膜識別的虹膜定位方法，並解決訓練圖片數量不足的問題，最後使用卷積神經網路(convolutional neural network，CNN)進行特徵學習及分類，準確度達到99.6%。

關鍵字

虹膜識別 ； 卷積神經網路 ； YOLO

並列摘要

Taiwan is a country that is passionate about pigeon racing. Pigeon competition often bring huge bonuses and bets, which induces many unscrupulous persons to attempt to use the methods of cheating and stealing pigeons to obtain benefits illegally, resulting in endless fairness and legal problems. Among them, the problem of the identity verification of racing pigeons has been criticized for a long time. In the past, most of the racing pigeon associations relied on the traditional foot rings for identity verification of the pigeons, which is inefficiency and unreliability. Although it has been added radio frequency identification (RFID) technology to improve the electronic foot rings, this encryption method still has security concerns that have been cracked. In recent years, there has also been an industry for pigeon DNA testing. It can achieve very high accuracy, but it is expensive and cannot be compared instantly. Therefore, the purpose of this paper is to propose a pigeon identification system with lower cost, faster verification ability and high accuracy. Based on the hypothesis, the pigeon iris is as unique and life-changing as the human, this study considers that pigeon iris recognition is a method to accurately verify the identity of a pigeon. It is more difficult to obtain pigeon iris images than humans, so few related researches on pigeon iris recognition at present. The contribution of this paper is to automatically detect the pigeon iris area through the improved YOLO object detection algorithm, replacing the traditional iris localization method, and solving the problem of insufficient training pictures. Finally, the convolutional neural network (CNN) is used for feature learning and classification, the accuracy reached 99.6%.

並列關鍵字

Iris recognition ； convolutional neural network ； YOLO

參考文獻

陳敬瑜、周家德、石勝文（2009）。鴿子虹膜識別系統，NCS2009。

廖士傑（2015）。如何由眼睛外觀自我診斷身體疾病。早安健康

賴寧寧（2011）。勝負一秒間。商業週刊，1227，臺北市

Alvarez-Betancourt, Y., & Garcia-Silvente, M. (2016). A keypoints-based feature extraction method for iris recognition under variable image quality conditions. Knowledge-Based Systems, 92, 169-182.

Aparna, S. (2017). Biotechnology market analysis by application (health, food and agriculture, natural resources and environment, industrial processing bioinformatics), by technology and segment forecasts, 2014-2025. Grand View Research, San Francisco, available at: www. grandviewresearch. com/industry-analysis/biotechnology-market.