虹膜紋理定位方法實作與比較

Implementation and Comparison of Iris Pattern Localization Methods

陳威榤(Wei-Chieh Chen)

指導教授 ： 石勝文

國立暨南國際大學/科技學院/資訊工程學系/碩士(2022年)

https://doi.org/10.6837/ncnu202200264

摘要

虹膜辨識因為它的穩定性、 非接觸式、 不易竊取、 高分辨率等特性， 這二十幾年來虹膜辨識一直是最受大家關注的生物辨識技術。 在虹膜辨識計算流程中， 虹膜邊界定位精確度對辨識精確度有很大的影響。 為瞭解使用何種技術可達成較好的定位結果， 本論文實現 Daugman 的 IDO (Integral Differential Operator) 以及基於 RANSAC (Random Sample Consensus) 的 IDO360-L, IDO180-L 以及 RANSAC-L 三種虹膜外緣 (Limbus) 定位，以及 IDO-E, RANSAC-E w/ ER, 與 RANSAC-E w/o ER 等三種上下眼皮邊界的定位方法。 其中 IDO360-L 與 IDO180-L 主要的差異是後者在計算 IDO 時， 只採左右兩側各90度的範圍以避開上下眼皮區間。 RANSAC-E w /ER 是在偵測眼皮邊界前， 先執行睫毛移除 (Eyelash Removal, 簡寫為 ER)， 而 RANSAC-E w/o ER 則無。 評比不同方法的準確度與運算效率時是採用本實驗室先前由 248 隻眼睛取得的 39,560 張近紅外線眼睛影像。 這些影像的虹膜外緣與上下眼皮邊界經過多次人工檢查與調整， 可以作為虹膜定位的參考值。 判斷定位精確度的方法，則是採用自動定位與人工定位結果的 IoU (Intersection over Union) 值。 實驗結果顯示，在定位虹膜外緣與上下眼皮邊界的精確度皆優於基於 IDO 的方法。 不過基於 RANSAC 的方法所需計算時間通常是基於IDO的方法的10--22 倍之間。 而在虹膜外緣的定位時， IDO180-L 精確度優於 IDO360-L。 最後， 我們也測試不同的定位方法會不會影響虹膜辨識率， 實驗結果顯示辨識正確率與定位精確成正比。

關鍵字

虹膜取像 ； 虹膜切割 ； 積分微分方程 ； 隨機抽樣一致性

並列摘要

Iris recognition has been the biometric recognition technology that has attracted the most attention in the past two decades because of its stability, non-contact, difficult to counterfeit, and high discrimination. In the computation process of iris identification, the localization accuracy of the iris boundaries has strong impact on the iris recognition accuracy. In order to understand which technique can achieve better localization results, we implement two variants of Daugman's IDO (Integral Differential Operator) limbus localization method, namely, IDO360-L and IDO180-L. The main difference between IDO360-L and IDO180-L is that the latter only uses 90-degree ranges on both the left and right sides of the limbus to avoid image areas in the upper and lower eyelids when calculating IDO. For comparison, we implement a RANSAC (Random Sample Consensus)-based limbus localization method, called RANSAC-L. As for eyelid boundary localization, we implement an IDO-based method, called IDO-E, and two RANSAC-based methods, i.e., RANSAC-E w/ ER and RANSAC-E w/o ER. RANSAC-E w/ ER performs eyelash removal before RANSAC eyelid detection, while RANSAC-E w/o ER does not. The accuracy and computational efficiency of the different methods were evaluated using 39,560 near-infrared eye images previously obtained from 248 eyes in our laboratory. The limbus of the iris and the upper and lower eyelid boundaries of these images have been manually checked and adjusted for many times. Therefore, the manual localization results can be used to assess the localization methods with IoU (Intersection over Union) metric. Experiment results show that the accuracy of localizing the iris boundaries is better than that of the IDO-based method. However, computation time of the RANSAC-based methods are usually 10--22 times longer than that of IDO-based methods. Furthermore, the IDO180-L is more accurate than the IDO360-L in the localization of the iris limbus. Finally, we also test whether different localization methods will affect the iris recognition rate. Iris recognition results show that the recognition accuracy is proportional to the iris localization accuracy.

並列關鍵字

Iris Image Acquisition ； Iris Segmentation ； Integro-Differential Equation ； Random Sample Consensus

參考文獻

[1] R. P. Wildes, “Iris recognition: an emerging biometric technology,” Proceedings of the IEEE, vol. 85, no. 9, pp. 1348–1363, 1997.

[2] T. v. d. Putte and J. Keuning, “Biometrical fingerprint recognition: don＇t get your fingers burned,” in Smart Card Research and Advanced Applications. Springer, 2000, pp. 289–303.

[3] H. Kang, B. Lee, H. Kim, D. Shin, and J. Kim, “A study on performance evaluation of the liveness detection for various fingerprint sensor modules,” in International Conference on Knowledge-Based and Intelligent Information and Engineering Systems. Springer, 2003, pp. 1245–1253.

[4] E. Marasco and A. Ross, “A survey on antispoofing schemes for fingerprint recognition systems,” ACM Computing Surveys (CSUR), vol. 47, no. 2, pp. 1–36, 2014.

[5] W. Wu, S. J. Elliott, S. Lin, S. Sun, and Y. Tang, “Review of palm vein recognition,” IET Biometrics, vol. 9, no. 1, pp. 1–10, 2020.