劉小飛 李明杰

摘要：對(duì)于不同傳感器不同時(shí)間段獲取的多幅低質(zhì)量虹膜圖像，采用圖像融合技術(shù)對(duì)其進(jìn)行處理。根據(jù)虹膜圖像瞳孔縮放非剛性形變的特點(diǎn)，提出用曲線擬合的方法進(jìn)行虹膜圖像的配準(zhǔn)，將配準(zhǔn)后的虹膜圖像利用像素級(jí)的小波融合方法進(jìn)行融合，最終目的是將多幅缺失信息的低質(zhì)量虹膜圖像融合成一幅含有更為豐富信息的虹膜圖像，并以此構(gòu)建新的虹膜圖像庫(kù)用于虹膜識(shí)別領(lǐng)域。實(shí)驗(yàn)表明，該方法使低質(zhì)量虹膜圖像在識(shí)別精度上得到了進(jìn)一步的提升。

關(guān)鍵詞：低質(zhì)量虹膜圖像；曲線擬合；小波融合；虹膜識(shí)別

中圖分類號(hào)：TP391.41文獻(xiàn)標(biāo)識(shí)碼：A文章編號(hào)：1009-3044(2012)20-4965-05

The Research of Low Quality Image Recognition Method Based on Iris Images

LIU Xiao-fei, LI Ming-jie

（Sanya University,Sanya 572022,China）

Abstract: For lots of low quality iris images which obtained from different sensors at different periods of time, using the image fusion tech? nology to process it. According to the non-rigid deformation characteristics of iris image zoom pupil, proposed using curve fitting technol? ogy in iris image registration.Then the registered iris image using pixel level fusion method of wavelet fusion for fusion,the ultimate aim is to make the missing pieces information of low quality iris images fused into an iris image which contains richer informations,and build a new iris image database for the field of iris recognition..The experimental results show that this method has futrher promoted the accuracy of low quality iris images recognition.

Key words: low-quality iris images; curve fitting; wavelet fusion; iris recognition

傳統(tǒng)的虹膜識(shí)別系統(tǒng)中，對(duì)于低質(zhì)量的虹膜圖像的處理，是建立圖像質(zhì)量評(píng)估標(biāo)準(zhǔn)，對(duì)符合標(biāo)準(zhǔn)的虹膜圖像認(rèn)為其質(zhì)量較高，予以保留，而沒有達(dá)到標(biāo)準(zhǔn)的虹膜圖像就被舍棄，很顯然在傳統(tǒng)的虹膜識(shí)別系統(tǒng)中并未考慮低質(zhì)量虹膜圖像的識(shí)別問(wèn)題。

該文提出將圖像融合技術(shù)應(yīng)用于低質(zhì)量虹膜圖像的拼接中來(lái)，提高低質(zhì)量虹膜圖像的質(zhì)量，使其可以用于身份識(shí)別。和傳統(tǒng)可以使用圖像融合技術(shù)的圖像相比，虹膜圖像融合的難點(diǎn)在于虹膜圖像的形變是非剛性的，不能用傳統(tǒng)的變換矩陣進(jìn)行配準(zhǔn)，該文采用了曲線擬合方法進(jìn)行配準(zhǔn)，對(duì)配準(zhǔn)后的虹膜圖像采用像素級(jí)融合，得到結(jié)果圖像后用于識(shí)別。

(a)為參考圖像，(b)待配準(zhǔn)圖像，(c)為高斯曲線擬合重構(gòu)圖像，(d)為拋物線擬合重構(gòu)圖像，(e)為(a)(c)融合后的圖像，(f)為(a)和(d)融合后的圖像。

圖6低質(zhì)量虹膜圖像融合2.3融合虹膜圖像效果的評(píng)價(jià)

該文采用相關(guān)系數(shù)[8]的評(píng)價(jià)方法。相關(guān)系數(shù)用來(lái)表示兩幅圖像的相關(guān)程度，通過(guò)相關(guān)系數(shù)我們可以得到測(cè)試圖像和融合圖像的相關(guān)程度。相關(guān)系數(shù)的值越接近于1，說(shuō)明兩幅圖像的相關(guān)程度越高。相關(guān)系數(shù)由下式來(lái)定義：

系數(shù)進(jìn)行評(píng)價(jià)。分別對(duì)3組不同人眼的低質(zhì)量虹膜圖像測(cè)試樣本進(jìn)行對(duì)比實(shí)驗(yàn)，計(jì)算出了每組對(duì)應(yīng)的相關(guān)參數(shù)。結(jié)果表明：該文提出的方法能夠有效的改變低質(zhì)量虹膜圖像識(shí)別的準(zhǔn)確度。

[1] Daniel Walsh,Adrian E.Raftery.A novel Hough transform based on eliminating particle swarm optimization and its applications[J]. Pattern Recognition, 2009,42(9):1959-1969.

[2]王葳,唐一平,任娟莉.一種改進(jìn)的Harris角點(diǎn)提取算法[J].光學(xué)精密工程,2008,16(10):1995-2001.

[3] Guiyun T,Duke G,David T. Comprehensive interest points based imaging mosaic[J].Pattern Recognit--ion Letters,2003,24(9-10): 1171-1179.

[4] Schmid C.,Mohr R.Local gray-value invariants forimage retrieval.IEEE Trans.Pattern Anal.Machine Intell, 1997,19(5):530-535.

[5] Yuan-Hai Shao, Nai-Yang Deng,Zhi-Min Yang.Least squares recursive projection twin support vectormachine for classification. Pattern Recognition,2012,45(6):2299-2307.

[6]白建軍,陳其松,張欣,楊彥龍.基于形態(tài)濾波的小波融合圖像增強(qiáng)算法[J].計(jì)算機(jī)仿真,2012,29(1):264-268.

[7]王琛,漆德寧,儲(chǔ)彬彬,龐璐璐.基于小波變換的圖像融合方法研究[J].電腦知識(shí)與技術(shù),2009,5(35):10075-10076.

[8]呂林濤,石富旬.漸進(jìn)式虹膜圖像質(zhì)量評(píng)估模型[J].計(jì)算機(jī)工程,2010,36(18):217-219.

[9] Chinese Academy of Sciences:Instituteof Automat--ion, Database of 756 Greyscale Eye Images[DB],＜http://www.sinobiometrics.com＞Version 1.0, 2003.