2023年全國碩士研究生考試考研英語一試題真題(含答案詳解+作文范文)_第1頁
已閱讀1頁,還剩16頁未讀, 繼續(xù)免費閱讀

下載本文檔

版權(quán)說明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請進(jìn)行舉報或認(rèn)領(lǐng)

文檔簡介

1、<p><b>  中文4590字</b></p><p>  出處:Procedia Technology 19 ( 2015 ) 820 – 826</p><p>  本科畢業(yè)設(shè)計外文翻譯</p><p><b> ?。ˋndroid)</b></p><p>  計算機(jī)科學(xué)與技術(shù)

2、學(xué)院 計算機(jī)科學(xué)與技術(shù) 專業(yè)2011年級4班</p><p>  Procedia技術(shù)</p><p>  可登陸網(wǎng)站:www.sciencedirect.com</p><p>  Procedia技術(shù)19 (2015) 820 – 826</p><p>  第八屆國際多科性跨領(lǐng)域研討會于2014年10月9-10日在羅馬尼亞的特爾古穆列召

3、開</p><p>  在安卓平臺的擊鍵動力學(xué)</p><p>  Tirgu Mures 在Sapientia大學(xué)技術(shù)學(xué)院和人文科學(xué)院發(fā)表, Soseaua Sighisoarei 1C 540485,羅馬尼亞</p><p><b>  摘要</b></p><p>  現(xiàn)在人們將越來越多的私人數(shù)據(jù)存儲在他們的移動設(shè)

4、備內(nèi),因此,加強(qiáng)現(xiàn)有的身份驗證機(jī)制是非常重要的。這種分析模式在專業(yè)領(lǐng)域內(nèi)稱作按鍵動態(tài),主要用于加強(qiáng)輸入密碼的安全性能。而且,觸屏還添加了其他功能特征:范圍從屏幕壓力或鍵盤對于經(jīng)典的時效性都被應(yīng)用于擊鍵動力學(xué),在這個論文里面,我們檢驗這些添加的屏幕功能對身份驗證機(jī)制的功能作用,并且通過驗證我們的42個用戶的數(shù)據(jù)庫的運轉(zhuǎn)。結(jié)果顯示這些添加的功能加強(qiáng)了雙重進(jìn)程。</p><p>  作者Elsevier Ltd 2

5、015發(fā)表年 本文對持有CC BY-NC-ND 證件的用戶是免費開放的。</p><p>  (http://c reatwecommons.org/licenses/b y -nc-ncl/4.0/).</p><p>  本文通過特爾古穆列大學(xué)工程系Petru Maior的雙重審核</p><p>  關(guān)鍵詞:安全性,用戶鑒別,按鍵動態(tài),觸摸特性</p&

6、gt;<p><b>  1.引言</b></p><p>  現(xiàn)在,越來越多的人將私人信息以及敏感信息存入智能手機(jī),因此,手機(jī)安全身份驗證方法的需求量日趨增長。用戶輸入口令是用戶防止設(shè)備入侵最常用的方法。然而,人們趨于使用口令操作,這樣用于方便記憶,易于開機(jī),因此,額外的機(jī)制需要用于身份驗證以加強(qiáng)口令安全。這樣一個互補(bǔ)的方法被應(yīng)用于用戶的輸入模式的稱之為擊鍵動力學(xué)。擊鍵動力學(xué)

7、在臺式電腦研究中是一個熱門的研究課題,而有研究手機(jī)的課題很少,觸屏智能手機(jī)的課題就更少。這項研究的最主要問題是觸屏是否可以添加新的功能特性,例如按壓或手指區(qū)域可以增強(qiáng)按鍵身份系統(tǒng)的精確性。下一節(jié)簡要介紹了擊鍵動力學(xué)在觸摸屏設(shè)備上的研究領(lǐng)域,以及回顧課題。接著我們通過數(shù)據(jù)的收集并評估再提出研究方法。最后一節(jié)提出了幾條結(jié)論和未來研究和發(fā)展的方向。</p><p><b>  2.擊鍵動力學(xué)</b>

8、;</p><p>  擊鍵動力學(xué)是一個熱門的研究領(lǐng)域,最重要的優(yōu)勢之一是成本低并且安裝簡便,與其他生物識別方法相比,該方法不需要任何專門的硬件設(shè)備[11]。因為作為掌控?fù)翩I模式是應(yīng)用一個后端軟件。這使得該方法可以為用戶的透明和非侵入性發(fā)揮功能,基建動力學(xué)可以被用于加強(qiáng)輸入密碼時的身份驗證和連續(xù)的身份驗證機(jī)制[2]。與其他的方式相比較,這種生物識別的精確度并不高[11]。擊鍵動力學(xué),擊鍵力學(xué)的研究報告收錄了各種輸

9、入設(shè)備的數(shù)據(jù),從常規(guī)的到壓力敏感的鍵盤,最常用時效性能是間歇時間和運轉(zhuǎn)時間。間歇時間是按鍵和輸出之間的時間間隔(又是被稱作停歇功能),而運行時間是上一個字符的輸出和下一個字符輸入之間時間。常用的功能(n-圖)是應(yīng)用三個或更多的連續(xù)擊鍵時間,但是大部分論文采用有向圖(兩個連續(xù)的擊鍵)。大多數(shù)現(xiàn)有的識別方法是檢測按鍵識別性能,包括分析研究數(shù)據(jù)和機(jī)器學(xué)習(xí)的方法。最簡單的方法是為一個資歷較深的用戶構(gòu)建一個模板 ,然后計算在實際的身份驗證階段輸

10、入模式和參考模板之間的距離,這種方法稱作樣板匹配,該方法還可以結(jié)合不同的衡量值,范圍可從簡單的歐式距離到馬氏距離。其中神經(jīng)網(wǎng)絡(luò)和無線電導(dǎo)航機(jī)是最好的[11]。</p><p>  生物識別性有兩個最突出的功能:驗證和確認(rèn),驗證過程是一個二元判定問題,在該程序系統(tǒng)接受或拒絕用戶的身份宣稱,身份鑒定也稱之為身份識別屬于一個分類問題: 系統(tǒng)將輸入模式分為N已知的一類。</p><p>  生物識

11、別系統(tǒng)假廢品率(FRR)是生物系統(tǒng)的錯誤拒絕真正的用戶提供的樣本種錯誤分別是:FAR,FRR,ERR。錯誤接受率(FAR)是生物識別系統(tǒng)在識別錯誤地評判并接受一個入侵者的信息;錯誤拒絕率(FRR)是指真正用戶的信息遭到生物識別系統(tǒng)的拒絕;ERR是指錯誤率同錯誤拒絕率相等的情況。</p><p>  下面的研究概述是關(guān)于觸摸屏的初級設(shè)備被用于數(shù)據(jù)收集。</p><p>  Saevanee

12、和 Bhattarakosol已經(jīng)陳述了初級研究學(xué)習(xí)用擊鍵動力學(xué)結(jié)合手指點擊力度。根據(jù)從10個用戶那里收集的數(shù)據(jù)集,他們驗證了可通過用戶只使用手指識別信息但準(zhǔn)確率高達(dá)99%。</p><p>  然而,觸屏版筆記本可用于收集數(shù)據(jù),由于不同的用戶有不同的手機(jī)號碼,所以不同的用戶必須輸入10個不同的長串?dāng)?shù)字,對于錯誤拒絕率(FRR)可以用數(shù)據(jù)庫評判,而錯誤接收率的錯誤數(shù)據(jù)必須被收集,虛假數(shù)據(jù)的缺乏是本次研究的最主要

13、限制。</p><p>  另一項研究Johansen的碩士論文里面被陳述,過是擊鍵動態(tài)同觸屏性能相關(guān),研究的主要目的在于比較擊鍵動態(tài)在個人電腦和智能手機(jī)的區(qū)別,一共有42個人參與本次實驗,其中有一部分人既參與了個人電腦又參與了智能手機(jī)的體驗。這項研究的最主要發(fā)現(xiàn)是:在僅使用于時間特性的情況下智能手機(jī)的運行狀況比個人電腦要差一些,在智能機(jī)上的運行比一個標(biāo)準(zhǔn)的鍵盤要低。然而在使用智能機(jī)的附加功能(包括時間性能)時,

14、運行狀態(tài)明顯比在一個標(biāo)準(zhǔn)鍵盤上要好,該研究擬將用于解答模仿某人的打字節(jié)奏有多困難。研究表明,在標(biāo)準(zhǔn)鍵盤上模擬人打字要比智能機(jī)上更容易,這樣研究的主要限制在于在數(shù)據(jù)收集過程中要使用數(shù)字式的口令在手機(jī)的12個鍵上。</p><p>  Trojahn在論文里面陳述過這項研究的主要目的,即用來證明手指壓力和尺寸作為附加性能可用于降低用戶認(rèn)證系統(tǒng)中的錯誤率。測試中需要152個數(shù)據(jù)的提供者引入一串17個數(shù)字式的口令密碼。每

15、一個參與者需要敲擊10次單的語句音。將運行時間,有像圖同三線圖形和字母計時信息結(jié)合是最好的FAR+FRR組合,其錯誤率遠(yuǎn)低于使用觸摸屏的附加功能,該研究的最主要限制同Johansen's所陳述的相同,并且數(shù)據(jù)是呈單語句音的形式運行的。</p><p>  我們發(fā)現(xiàn)只有一項研究將手機(jī)軟鍵盤用于用戶身份驗證[4],輸入的數(shù)據(jù)是由13個用戶在3周的時間所收集的,高級的軟件鍵盤在一個普通軟件的輔助下可將關(guān)鍵的按鍵

16、信息存儲在同一文本中。按鍵長度、平滑度、壓力、手指區(qū)域和設(shè)備定位方向被用作功能特性,可通過使用使用遠(yuǎn)FAR和FRR報告用戶身份驗證結(jié)果。的數(shù)據(jù)的收集機(jī)制定義得并不是很清楚,觸摸和按鍵的概念區(qū)別將會在Draffin提出的方法中證明是有價值的。</p><p>  最近Sen和Muralidharan提出的一項研究使用了移動設(shè)備上的壓力作為用戶身份驗證的功能[10],同其他的研究相似,該實驗是基于一個4位數(shù)字的密碼。

17、</p><p>  除了確認(rèn)驗證結(jié)果使用了PAR和FRR類型錯誤,ERR錯誤也被報告是基于一個特殊的虛擬模型。</p><p>  表格1總結(jié)了這些最新的研究結(jié)果,遺憾的是不同的錯誤類型在統(tǒng)一表格中使比較工作變得棘手。</p><p>  表格1 最近的研究獲得了在觸摸屏上在運行的錯誤率 (性能將會在表2中進(jìn)行解釋)。</p><p> 

18、 結(jié)論,我們可以陳述的是沒有研究在智能機(jī)上使用真實的口令密碼,并且觸屏性能-壓力和手指區(qū)域沒有被在真實在條件中研究過。</p><p><b>  3.1方法論 </b></p><p>  每個Android應(yīng)用程序有自己相應(yīng)的軟件鍵盤,被用于開發(fā)數(shù)據(jù)收集系統(tǒng)。用戶必須輸入私人信息,如:性別,出生日期,關(guān)于在使用智能手機(jī)注冊階段的經(jīng)驗,因為輸入模式可以受到幾個因素的

19、影響,所以收集的數(shù)據(jù)應(yīng)該是幾段會話。大部分的實驗參與者在兩周內(nèi)都完成了2段會話,同一段會話必需由參與者重復(fù)30次,這被認(rèn)為是一個安全系數(shù)很高的密碼,也被應(yīng)用于Killourhy的擊鍵力學(xué)實驗設(shè)計[6]。共42人參加了這項研究,其中24名男性,18名女性24歲,年齡層從20-46(平均年齡在22、2歲),其中有一位老師其余都是學(xué)生。我們從收集到的數(shù)據(jù)包排除了含刪除和創(chuàng)建的數(shù)據(jù)集,數(shù)據(jù)包是從51個用戶的數(shù)據(jù)輸入模式里收集的。我們決定為每個用

20、戶設(shè)定相同的密碼,這樣每個參與者的語音數(shù)據(jù)都可用于分辨一個非法用戶和一個合法的用戶。</p><p>  采集的數(shù)據(jù)可供兩種類型的Android設(shè)備使用,一種是為7尺寸平板電腦和美孚LG的P710設(shè)備。總共有37個平板電腦用戶和5手機(jī)用戶提供數(shù)據(jù)。</p><p>  輸入所選的口令密碼鍵盤上需要14鍵,8個字母, 一個數(shù)字,一個字符,點擊兩次Shift鍵實現(xiàn)從大寫字母到數(shù)字的切換。觸摸屏

21、幕即可運行節(jié)省了使用手指敲擊的時間,釋放保存屏幕的時間戳,無線導(dǎo)航的組成性能可以參考表2。</p><p><b>  3.2 測量方法</b></p><p>  用戶信息驗證可以使用常用的學(xué)習(xí)軟件WEKA(3.6.11版)[13],顯示結(jié)果良好,結(jié)果的顯著差異是在于其使用了正確的配對,顯示值為0.05, Weka的搜索方法提到了一些默認(rèn)的參數(shù)的優(yōu)化分類器。<

22、/p><p>  各種分類器在擊鍵動力學(xué)數(shù)據(jù)集中使用廣泛,被使用之前:統(tǒng)計方法、決策樹、神經(jīng)網(wǎng)絡(luò)法、模糊法,支持向量法,這些在列表中都可查詢[11]。對于此項論文,我們從Weka的論文里面挑選出一些實際的方法,覆蓋了各類機(jī)器學(xué)習(xí)法。</p><p>  Naive Bayes的分類是基于Bayes'真理的概率分類法,這個分類法假設(shè)所有的特性都是獨立的一個實例但是分類法通常是不正確的。盡

23、管這種方法是幼稚的低級的,這種分類法適用在應(yīng)用程序使用廣泛。</p><p>  Bayesian的網(wǎng)絡(luò)是一種概率模型, 用一個有向無環(huán)圖表示一組隨機(jī)變量及其依賴條件的關(guān)系,圖的結(jié)點是表示隨機(jī)變量和幾個代表性的邊緣是有條件地依賴變量[1]。</p><p>  最近,行業(yè)同事Weka的研究(k-NN, IBk)是基于實例的分類算法,該算法是最新的。決策樹分發(fā)是非常受歡迎的方法,該方法像圖一

24、樣以樹為基礎(chǔ),近年來出現(xiàn)的算法都可用于擊鍵動力學(xué)。在許多應(yīng)用領(lǐng)域我們選擇使用Weka的J48 C4.5作為最佳方法。隨機(jī)森林分類器[3]是一個集成學(xué)習(xí)方法,該方法隨機(jī)引入決策樹結(jié)構(gòu)的一組(我們使用100棵樹)進(jìn)行評估。支持向量機(jī)可建立一個線性判別函數(shù)將分類實例進(jìn)行分類。如果沒有線性分離是可使用,內(nèi)核地圖可將實例映射到一個高維特征空間。我們通過使用Weka的內(nèi)項機(jī)核實現(xiàn)了LibSVM。C和y通過網(wǎng)絡(luò)搜索法被明顯優(yōu)化成兩個數(shù)據(jù)集(在41個特

25、性集里C = 10.55 y =1.86 ,在71個特性集里C = 7.46 y = 0.25),并且所有輸入功能都被正常且為(0 - 1)。</p><p>  多層感知器(MLP)是由反向傳播訓(xùn)練的人工神經(jīng)網(wǎng)絡(luò),我們這里只說明通過Weka提出的方法實現(xiàn)默認(rèn)設(shè)置的結(jié)果(數(shù)量的隱藏層是Weka的默認(rèn)設(shè)置(數(shù)的屬性+數(shù)量的分類)/ 2)。</p><p><b>  4. 結(jié)果&l

26、t;/b></p><p><b>  4.1識別結(jié)果</b></p><p>  為了顯示按鍵數(shù)據(jù)之間精度的差異分類有和沒有觸摸屏(壓力和手指區(qū))的特性,有兩個數(shù)據(jù)集被使用。其中第一個數(shù)據(jù)集包含41特性(H + DD +AH),第二個數(shù)據(jù)集包含71個特性(H + DD + UD+P+FA+AH+AP+AFA).</p><p>  我們

27、使用收集到的數(shù)據(jù)沒有經(jīng)過任何轉(zhuǎn)換,也沒有二次特性計算或特征選擇(除了標(biāo)準(zhǔn)化的支持向量方法)。沒有增加或調(diào)優(yōu)其他的方法使用,除了基于內(nèi)部隨機(jī)化的隨機(jī)森林分類器的使用,</p><p>  這項工作使用的數(shù)據(jù)可以參考網(wǎng)站http://www.ms.sapientia.ro/~manyi/keystroke.html.</p><p>  表3 分類精度測量的兩個數(shù)據(jù)集。第二列顯示了第一個數(shù)據(jù)集

28、的精度:41特性。</p><p>  第三列顯示了第二個數(shù)據(jù)集的精度:71的特性。</p><p>  表3給出了這些方法的精度。我們報告了分類精度10分的解釋,10分層交叉驗證對整個數(shù)據(jù)集(共100分,標(biāo)準(zhǔn)偏差在圓括號中)。所有情況明顯表現(xiàn)在71個特征數(shù)據(jù)集(在0.05顯著性水平)。</p><p><b>  4.2 驗證結(jié)果</b>&l

29、t;/p><p>  Killourhy和 Maxion R[8]提供的腳本]驗證了該測量方法,該腳本基于歐幾里得,曼哈頓提出的EER的三種異常計算檢測方法,試驗中數(shù)據(jù)歸一化,然后分區(qū)分成三個相等的部分,每個部分包含17個用戶的口令數(shù)據(jù)信息,三分之二的數(shù)據(jù)用來創(chuàng)建用戶模板而其余的三分之一用于測試FRR。前五個口令密碼來自每個用戶的數(shù)據(jù),除了其中一個數(shù)據(jù)是用作FAR測試。驗證需要重復(fù)3次,每次測量重復(fù)了3次,每次使用不

30、同的范圍為測試對象。表4總結(jié)了驗證結(jié)果,從三個測量可獲得的平均的EER,最低的誤差(12.9%),該數(shù)據(jù)是通過Manhattan基于時間和觸摸屏的特性獲得的數(shù)據(jù)。最低的EER是基于時間的性能為15.3%,該數(shù)據(jù)是Manhattan所提出的度量標(biāo)準(zhǔn)。</p><p>  結(jié)論,通過觸摸屏的功能不僅提高了分類精度也驗證了其精度。</p><p><b>  5.總結(jié)</b>

31、;</p><p>  在本文中,我們實驗性地證明了演示了以觸摸屏的特性為基礎(chǔ)的性能促進(jìn)了基于識別和驗證為基礎(chǔ)的擊鍵力學(xué)性能。數(shù)據(jù)集的收集可使用Android觸摸屏設(shè)備時間和觸摸屏的基本特性也進(jìn)行了研究。身份識別的驗證過程使用了幾個機(jī)器學(xué)習(xí)分類算法,其中隨機(jī)森林法,Bayesian的神經(jīng)網(wǎng)絡(luò)法和SVM是最好的分類算法。不僅在身份識別過程中,在驗證測量過程中也使用了相同的數(shù)集。在這種情況下EER在計算過程中使用了不

32、同的的距離度量,包括; Euclidean, Manhattan and Mahalanobis..其Manhattan是最好便于執(zhí)行的距離函數(shù)。</p><p>  在識別測量的情況下,將以觸摸屏為基礎(chǔ)的附加功能設(shè)置為默認(rèn)的特性,為每個分類器增加超過10%的精度,其錯誤率降低了24%( Manhattan度量標(biāo)準(zhǔn)),</p><p>  這個改進(jìn)是很難下驗證測量的情況觀察到的。</

33、p><p>  在數(shù)據(jù)預(yù)處理階段,我們觀察到一些鍵入模式包含刪除模式從數(shù)據(jù)集中刪除了。</p><p>  然而,這些誤差可以被認(rèn)為是用戶的一個單獨的功能,該功能的可以在后續(xù)階段進(jìn)行研究。</p><p><b>  鳴謝</b></p><p>  本研究獲得了Sapientia(科學(xué)研究所)的大力支持</p>

34、<p><b>  參考文獻(xiàn)</b></p><p>  [1]Bouckaert R R.在Weka的文獻(xiàn)中發(fā)表的網(wǎng)絡(luò)分類研究 2008。</p><p>  可參考網(wǎng)站http://www.cs.waikato,ac.nz/~remco/weka.bn.pdf</p><p>  [2] Bours P在《信息安全技術(shù)》報告中

35、發(fā)表的連續(xù)擊鍵動力學(xué):從不同的角度對生物識別性評估。</p><p>  《信息安全技術(shù)》報告第17卷,l-2節(jié),36-43頁。2012年2月</p><p>  [3] Breiman. L的.隨機(jī)森林法和機(jī)器學(xué)習(xí)法45節(jié)第一張p5-32。2001年。</p><p>  [4]Draffin B, Zhu J, Zhang J, KeySens 通過微小的建模

36、軟件對用戶進(jìn)行身份驗證。</p><p>  計算機(jī)科學(xué)研究所所發(fā)表的《社會信息和電信工程》 ,2014年 第130卷p.184-201 </p><p>  [5] Johansen UA的碩士論文,《擊鍵力學(xué)與觸摸屏設(shè)備》,2012年發(fā)表于Gjovik大學(xué)的計算機(jī)科學(xué)和媒體部門出版社。</p><p>  [6]在IEEE / IFIP討論的可靠的系統(tǒng)的網(wǎng)絡(luò)國

37、際會議中將擊鍵力學(xué)的異常檢測算法進(jìn)行比較, (DSN ' 09),2009年,p . 125 – 134</p><p>  [7] 在伊斯坦布爾,土耳其所召開的計算國際會議報告中, 《身份驗證系統(tǒng)使用擊鍵力學(xué)感應(yīng)壓力》進(jìn)行探討報告2004, p. 19-22</p><p>  [8]參考網(wǎng)站http://www.r-project.org/</p><p&g

38、t;  [9] Saevanee H和Bhattarakosol P在第6屆IEEE消費者通訊和網(wǎng)絡(luò)會議上探討《擊鍵力學(xué)和手指按壓驗證用戶身份信息》進(jìn)行探討2009, p. 1-2。</p><p>  [10] Sen S和 Muralidharan K. 第七次國際會議移動計算和無處不在的網(wǎng)絡(luò)(ICMU)所發(fā)表的《擊鍵在移動手機(jī)上的驗證性能》, 2014, p. 56-61.</p><p

39、>  [11] Teh P S, Teoh A B J和Yue S.關(guān)于擊鍵力學(xué)和生物識別技術(shù)的調(diào)查,發(fā)表在《科學(xué)世界》2013年,卷,ID 408280條, 24頁</p><p>  [12] Rojahn M, Amdt F和Ortmeier F.在2013年所發(fā)表的《觸屏鍵盤同n圖組合的影響》中關(guān)于按鍵動力學(xué)和用戶驗證信息。第三屆在國際會議上所探討的移動服務(wù),移動資源,移動用戶。(2013, p.

40、114-119)。</p><p>  [13] 由Witten I H, Frank E, Hall M, Data Mining,2011在Morgan Kaufmann所發(fā)表的《實用機(jī)器學(xué)習(xí)工具和技術(shù)》。</p><p>  Available online at www.sciencedirect.com</p><p>  ScienceDirect<

41、;/p><p>  Procedia Technology 19 (2015) 820 - 826</p><p><b>  Procedia</b></p><p>  Technologg</p><p>  8th International Conference Interdisciplinarity in Eng

42、ineering, INTER-ENG 2014, 9-10 0ctober</p><p>  2014:vTirgu-i/lures, Romania</p><p>  Keystroke dynamics on Android platform</p><p>  Margit Antal*, Laszlo Zsolt Szabo, Izabella Las

43、zlo</p><p>  Sapientia University, Faculty of Technical and Human Sciences, Soseaua Sighisoarei 10, Tirgu Mures (Corunca) 540485, Romania</p><p><b>  Abstract</b></p><p>

44、;  Currently people store more and more sensitive data on their mobile devices. Therefore it is highly important to strengthen the existing authentication mechanisms. The analysis of typing patterns, formally known as k

45、eystroke dynamics is useful to enhance the security ofpassword-based authentication. Moreover, touchscreen allows adding features ranging from pressure ofthe screen or finger area to the classical time-based features us

46、ed for keystroke dynamics. In this paper we examine the effect </p><p>  @ 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license</p><p>  (ht

47、tp://c reatwecommons.org/licenses/b y -nc-ncl/4.0/).</p><p>  Peer-review under responsibility of "Petru Maior" University of Tirgu Mures, Faculty of Engineering</p><p>  Keywords: Sec

48、urity; User Authentication; Behavioral Biometric; Keystroke Dynamics; Touch Features</p><p>  1. Introduction</p><p>  At present more and more people store private and sensitive data on their s

49、martphones. Consequently, the demand is growing for secure mobile authentication methods. Setting a password-based authentication is the most frequently used method to protect data from intruders. However, people tend to

50、 use passwords, which can be easily remembered, hence easy to crack. Therefore, additional mechanisms are needed to enhance the security of password based authentication. One such complementary method is t</p><

51、;p>  Keystroke dynamics is an active research topic and has been researched mainly on desktop computers. There are very few studies conducted on mobile phones, even fewer on smartphones with touchscreen. The main res

52、earch * Corresponding author. Tel.: +40 265 250 620; fax: +40 265 206 211</p><p>  E-mail address: manyi@rus.sapientia.ro.</p><p>  2212-0173@ 2015 The Authors. Published by Elsevier Ltd. This i

53、s an open access article under the CC BY-NC-ND license</p><p>  (http://creativecommons.org/l/censes/b y-nc-ncU4.0/).</p><p>  Peer-review under responsibility of "Petru Maior" Univers

54、ity of Tirgu Mures, Faculty of Engineering</p><p>  doi:10.1016/j .protcy.2015.02.1 18</p><p>  Margit Antal et a/. / Procedia Technology /9 (2015) 820 - 826</p><p>  question of t

55、his is study is whether new features provided by touchscreens - such as pressure or finger area - can</p><p>  improve the accuracy of a keystroke based authentication system.</p><p>  The next

56、section briefly presents the research field of keystroke dynamics, reviewing the research studies conducted on devices with touchscreens. Then we present the research methodology, including data collection and its evalua

57、tion through identification and verification measurements. The final section presents some conclusions and future directions.</p><p>  2. Keystroke Dynamics</p><p>  Keystroke dynamics is a heav

58、ily researched field. One of the most important advantages is low implementation and deployment cost [11]. In contrast to other biometric methods, this method does not require any dedicated hardware device. As the captur

59、e ofkeystroke pattern is implemented using a backend software, it makes this method transparent and noninvasive for the user [11]. Keystroke dynamics can be used both for strengthening entry point</p><p>  b

60、ased authentication and as a continuous authentication mechanism [2]. Compared to other methods, the main disadvantage of this type ofbiometrics is low accuracy [11].</p><p>  Keystroke dynamics studies repo

61、rted data acquisition using various input devices, ranging from normal to pressure sensitive keyboards [7]. The most commonly used time-based features are dwell time and flight time. Dwell time is the time interval betwe

62、en key press and key release (sometimes called hold time) whereas flight time is the time interval between releasing one key and pressing the next one. Sometimes three or more consecutive key</p><p>  time e

63、vents are used as features (n-graph), but the majority of papers used digraph features (two consecutive keys). Most of the existing pattern recognition approaches were tested for keystroke recognition, including statisti

64、cal and machine learning approaches. The simplest method is to construct a reference template for the respective user and compute the distance between the current typing pattern and the reference template in the authenti

65、cation stage. This method is known as template matchin</p><p>  Biometric systems can have two distinct functions: verification and identification. Verification is a binary decision problem, in which the sys

66、tem accepts or rejects the identity claimed by the user. Identification, also called recognition, is a classification problem: the system classifies the input pattern into one of the N known classes.</p><p>

67、  The quality of biometric systems is usually characterized by three kinds of errors: FAR, FRR and EER. False Acceptance Rate (FAR) is the rate at which a biometric system accepts a sample as one belonging to the claimed

68、 identity when the sample belongs to an impostor. False Rejection Rate (FRR) is the rate at which a biometric system incorrectly rejects a sample provided by the genuine user. EER is the rate at which FAR is equal to FRR

69、.</p><p>  The following is an overview of studies that used touchscreen based devices for data collection.</p><p>  Saevanee and Bhattarakosol presented the first study [9] using keystroke dyna

70、mics combined with finger pressure. Through a dataset collected from 10 users they demonstrated that users can be identified with 99% accuracy by using only finger pressure information. However, data were collected using

71、 a notebook with touchpad acting as a touchscreen. Participants had to enter their 10 digits long cell phone numbers. Since each user has a</p><p>  different phone number, only FRR type error can be measure

72、d on the dataset. For FAR error measurement impostor data must be collected. The lack ofimpostor data can be considered the main limitation of this study.</p><p>  Another study related to keystroke dynamics

73、 using touchscreen features is presented in the master thesis of Johansen [5]. The purpose of this study was to compare keystroke dynamics on personal computer to smartphones. A total of 42 persons took part in the exper

74、iment. Some of them completed both the personal computer and the smartphone experiment. The main finding of the study is that using only the timing features, the performance on smartphones is worse than on standard keybo

75、ard. However, usi</p><p>  The main goal of the study presented by Trojahn et al. in the paper [12] is to demonstrate that pressure and size of the finger as additional features reduce the error rate of a ke

76、ystroke-based authentication system. The test required each of the 152 data providers to introduce a 17-digit passphrase. Each participant typed the password ten times in a Margit Antal et a/. / Procedia Technology / 9 (

77、2015) 820 – 826 single session. The best FAR+FRR combination was obtained by using duration (hold ti</p><p>  We have found only one study using mobile soft keyboard for user authentication [4]. Typing data

78、was collected from 13 users during a period of 3 weeks. The developed soft keyboard collected key press information in all contexts requiring a soft keyboard. Key press length, drift, pressure, finger area and device ori

79、entation were used as features. User authentication results are reported using FAR and FRR errors. The data collection mechanism is not defined very clearly - explaining differences b</p><p>  A recent study

80、 using pressure as feature for user authentication on mobile devices is presented by Sen and Muralidharan [10]. Similar to other studies this one is based on a 4-digit password. Besides the presented verification results

81、 using FAR and FRR type errors, EER is also reported based on a special impostor model.</p><p>  Table l summarizes the results obtained by these recent studies. Unfortunately, the different types oferrors r

82、eported by the studies make the comparison difficult.</p><p>  Table l. Error rates obtained by recent studies conducted on touchscreen (Feature notations are explained in Table 2)</p><p>  In c

83、onclusion, we can state that no study has been on smartphones using a real-life password. Moreover, has not been studied in such a realistic environment.</p><p>  3. Methodology 3./. Data Collection designed

84、 to measure the performance of keystroke dynamics the effect of touchscreen features - pressure and finger area -</p><p>  An Android application having its own software keyboard was developed for data coll

85、ection (see Fig.l. (b)-(d》. Users had to introduce some personal data, such as gender, birth date and their experience level regarding smartphone usage in the registration phase (see Fig.l. (a》. Because typing pattern ca

86、n be influenced by several factors, data should be collected in several sessions. The majority of participants completed 2 sessions in a period of two weeks. In each session users had to enter the </p><p>  

87、number of 42 people took part in this study, 24 male and 18 female participants, aged 20-46 (with the average of 22.2 years). All the participants were students, except for one female teacher. We excluded from the collec

88、ted data input patterns containing deletions and created a dataset for measurements containing 51 input patterns from each</p><p>  user. We decided to use the same password for each user so that each parti

89、cipant data may be used in the measurements both as an impostor and as a legitimate user.</p><p>  For data collection two types of Android devices were used, a Nexus 7 tablet and a Mobil LG Optimus L7 11<

90、;/p><p>  P710 device. Allin all 37 tablet users and 5 mobile phone users supplied data.</p><p>  Typing the chosen password on our software keyboard required pressing 14 keys: 8 letters, a digit,

91、 a period</p><p>  character, twice the Shift key in order to switch to/from capital letters and twice the numerical keyboard switch key. Touching the screen triggers a touch down event in the system that sa

92、ves timestamp, pressure and finger area. When releasing the screen the timestamp is saved. The components ofthe feature vector are shown in Table 2.</p><p>  3.2. Measurements</p><p>  User iden

93、tification measurements were performed using WEKA (version 3.6.11) [13], a popular machine learning software. Significant differences in results were determined using corrected paired t-test at 0.05 significance level. S

94、ome of the default parameters of the classifiers were optimized with Weka's search methods(these are always mentioned).</p><p>  Various class/fiers were used previously on keystroke dynamics datasets: s

95、tatistical methods, decision trees, neural networks, fuzzy methods, support vectors are some in the list, for review see [11]. For this paper we selected</p><p>  some well-known algorithms implemented in We

96、ka, covering various machine learning methods.</p><p>  Naive Bayes is a probabilistic classifier based on Bayes' theorem. This classifier assumes that all features of aninstance are independent which is

97、 usually not true. In spite of this naive approach, this classifier works well in many real-world applications.</p><p>  Bayesian network is a probabilistic model that represents a set of random variables a

98、nd their conditional dependencies using a directed acyclic graph. The nodes of the graph are the random variables and the edges represent conditionally dependent variables [1].</p><p>  Margit Antal et a/. /

99、 Procedia Technology / 9 (2015) 820 - 826</p><p>  Nearest neighbors (k-NN, IBk in Weka) is an instance based classification algorithm where a new instance label is decided by the K closest neighbors (K=l wa

100、s used, giving the best results in our tests). Decision trees are extremely popular methods based on tree like graphs, and appeared between the algorithms used for keystroke dynamics in recent years. Their training and

101、testing time is fast, and classification results are among the best methods for many application areas. We used Weka's J48 i</p><p>  Support vector machines build a linear discriminant function that sep

102、arates the instances of classes. If no linea separation is possible, a kernel maps the instances into a high-dimensional feature space. We used the LibSVM</p><p>  implementation through Weka with radial ba

溫馨提示

  • 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
  • 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
  • 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內(nèi)容里面會有圖紙預(yù)覽,若沒有圖紙預(yù)覽就沒有圖紙。
  • 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
  • 5. 眾賞文庫僅提供信息存儲空間,僅對用戶上傳內(nèi)容的表現(xiàn)方式做保護(hù)處理,對用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對任何下載內(nèi)容負(fù)責(zé)。
  • 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請與我們聯(lián)系,我們立即糾正。
  • 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時也不承擔(dān)用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。

最新文檔

評論

0/150

提交評論