以眼球追蹤儀探討文字驗證碼工作之認知過程

A study of cognitive process of text-based CAPTCHA tasks by eye-tracking data

馬瑞家(Jui-Chia Ma)

指導教授 ： 李英聯

朝陽科技大學/理工學院/工業工程與管理系/碩士(2012年)

摘要

摘要
驗證碼（Completely Automated Public Turing test to tell Computers and Humans Apart, CAPTCHA）是預防駭客利用自動機器人程式濫用網路服務資源的安全性措施，其中以文字類型驗證碼最為普遍，但文字類型驗證碼仍然可透過電腦程式辨識而被破解。因此，文字型驗證碼需加上干擾的設計使電腦不易辨識，但過度的干擾設計可能會增加使用者在辨識時視覺與認知的負荷。
由於現有OCR軟體可利用字典的功能來加以辨識其文字驗證碼，因此目前文字驗證碼的設計普遍為word（文字組合的單字）與non-word（無意義的字母組合）兩種來提供使用者做辨識。在認知的過程中發現我們對於word較為熟悉的字屬於由上而下的處理過程反應速度較為快速，相較於non-word視覺系統必須將這些刺激分析並組合成一個字元是屬於由下而上的處理過程則會相對較慢，基於上述的因素將word與non-word兩組為第一個實驗因子。先前研究比較過文字驗證碼的反應時間，發現其中較難辨識的文字驗證碼和過去occluded object recognition研究中所提出的理論有所關聯，其中遮蔽文字的前景物件為visible（看得見的）是屬於線條干擾，前景物件呈現反白為invisible（看不見的）屬於切割干擾，不是物件也不是反白則為incoherent（不一致的）屬於雜點干擾，以上三種干擾組合為第二個實驗因子。本研究為探討這些因子組合的認知過程，以眼球追蹤儀觀察受測者眼球追蹤資料，比較受測者在辨識不同組合驗證碼時，工作績效、視覺疲勞、工作負荷及認知過程的差異。
研究結果顯示種類incoherent顯著縮短反應的時間且錯誤率較低。此外，用字word組合有助於提高受測者的辨識績效，並能有效減少主觀視覺上的疲勞。invisible干擾設計則會影響每個字元的眼球追蹤資料，顯示此干擾容易增加視覺上的負荷。文字閱讀認知過程由下而上的處理方式不但會增加使用者辨識上的負荷，甚至造成使用者的困擾。建議往後文字驗證碼設計者可搭配種類incoherent與用字non-word組合，以及種類visible與用字word組合，除了能有效防止文字驗證碼遭受破解外，並能減少辨識時認知處理上的負荷，以提高使用者通過的機率。

關鍵字

認知過程 ； 眼球追蹤儀 ； 工作負荷 ； 眼球追蹤資料 ； 驗證碼

並列摘要

Abstract
CAPTCHA, the acronym of “Completely Automatic Public Turning test to tell Computers and Human Apart”, is kind of security measures for preventing hacker to overuse the internet service and resource. However, text-based CAPTCHA is the most common security measure but this way also could be deciphered by computer program. Therefore, text-based CAPTCHA must add the disturbance on the design to cause the computer program could not easy to recognize, but the excessive disturbance design will possibly increase the identification visual and cognition loading when user is trying to recognize something.
Currently, OCR software can tell text-based CAPTCHA by using dictionary function so that the general designs of text-based CAPTCHA are word and non-word for users’ identification. In the course of cognition, we find that the familiar words are in a process from top-down with rapid reaction speed. However, compared to non-word visual system, it must form these stimulated analyses into a character in the process from bottom-up and the speed is slow. Word and non-word are treated as the first experimental factor based on the above reasons. The previous research has compared with the reactive time of text-based CAPTCHA and finds the difficult text-based CAPTCHA is related with the theory proposed in the occluded object recognition research. The foreground objects of covered characters are visible and belong to Thread Noise. If the foreground object is the reverse type, it is invisible and belongs to Line Mask. If foreground object is neither the object nor reverse type, it is incoherent and belongs to Blot Mask. The above three kinds of interferences are treated as the second experimental factor. This research discusses the cognitive process of these factor groups and observes subjects’ eye-tracking data through the eye tracker as well as compares subjects’ difference among working performance, visual fatigue, workload and cognitive process in different groups’ text-based CAPTCHA.
The research results show incoherence cuts response time obviously and the error rate is lower. Moreover, word combination helps to improve the identification performance of the subject and is effective in reducing the fatigue in visual sense. However, invisible interference design can affect the eye-tracking data of each character, easily increasing the load of visual sense. The bottom-up processing in text reading and cognition will not only increase the load of users in identification, but even make trouble for users. We suggest that designers of text-based CAPTCHA collocate incoherence with non-word combination, and collocate visibility with word combination. It is effective in both preventing text-based CAPTCHA being cracked and reducing the load of identification in cognition processing so as to enhance users’ passing rate.

並列關鍵字

CAPTCHA ； eye tracker ； workload ； eye-tracking data ； cognitive process

參考文獻

伊彬、林演慶，(2006)，視覺影像處理之眼球運動相關研究探討，設計學報第11卷第4期。

徐志翔，(2010)，探討年齡對文字驗證碼工作之影響，碩士論文，朝陽科技大學工業工程與管理系。

蔡介立、顏妙琁、汪勁安，(2005)，眼球移動測量及在中文閱讀研究之應用，應用心理研究，第28期，91-104頁。

陳學志、賴惠德、邱發忠，(2010)，眼球追蹤技術在學習與教育上的應用，眼球追蹤與學習教育，第55期，第4卷，39-68頁。

陳弘庭，(2001)，模糊分群方法、語意變數、分群群數關係之研究─以市場區隔為例，碩士論文，國立成功大學 工業管理科學系。