應用特徵整合理論於文字驗證碼

Application of feature integration theory to improve the design of text CAPTCHA

劉彥良(Yan-Liang lue)

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(2013年)

摘要

多數系統都會透過一種驗證碼 （ 也 就 是 CAPTCHA, Completely Automated Public Turing test to tell Computers and Humans Apart）機制，來防止有心人士利用自動機器人程式濫用網路資源。在各種類型的驗證碼當中，以文字型態最為普遍，但該型態驗證碼仍可利用光學文字辨識(Optical Character Recognition, 簡稱 OCR)破解。因此多數驗證碼需加上干擾使自動程式不易辨識，但這樣的設計同時也增加了人類辨識的難度。
為解決此問題，本研究提出以特徵整合理論(feature integration theory, 簡稱 FIT)改良文字型驗證碼的構想。在該理論中提到，人類認知系統在搜索preattentive features 時非常快，幾乎不需要注意力資源且搜索速度不受干擾物的多寡影響。本研究挑選三種符合 preattentive features 分別為大小、形狀、及線條融入常見的文字驗證碼干擾設計中，如字體扭曲、變形、額外線條等，將這些干擾設計加以改良。
本研究進行兩階段實驗，第一階段為確認三種符合 preattentive features
的設計是否符合 preattentive search 特性，即工作績效不因干擾物增加而受影響。第二階段實驗則是將通過第一階段實驗之驗證碼與常見的文字型態
與無干擾型態驗證碼進行比較。
研究結果顯示，第一階段實驗的線條設計並不符合 preattentive search 特性，因線條設計的目標物與干擾角度沒有太大差異。第二階段實驗結果顯示，改良過後的驗證碼能有效降低人類辨識難度與遭惡意程式破解，且比常見的驗證碼反應時間短、錯誤率低，主觀喜好程度較高。

關鍵字

前注意 ； 光學文字辨識 ； 特徵整合理論 ； 驗證碼

並列摘要

In order to avoid the malicious abuse of network resources, most of systems use a kind mechanism of CAPTCHA (that is CAPTCHA, Completely Automated Public Turing test to tell Computers and Humans Apart) to prevent an intentional personage from making use of automatical robot program to create accounts,vote,leave a message, or rate. Among the various types of CAPTCHA,the commonest type is literal,however,the text-based CAPTCHA could also be cracked by the OCR(Optical Character Recognition).Thus,the text-based CAPTCHA must be required with the interference design to obstruct computer identification ability, but this design will also increase the difficulty of human identification.

To solve this problem, this study proposes a concept of using the feature integration theory (FIT) to improve the text-based CAPTCHA. Mentioned in this theory, the human cognitive system is very fast in the preattentive features
search process, it hardly needs the attention resources and the search speed is not affected by the amount of the interferent.This research will select three kinds of preattentive features,such as size(differences in size as a design basis),shape(differences in shape design basis and line(differences in line as a design basis),which are integrated into the common text-based CAPTCHA interference design,respectively,and then improve the text-based CAPTCHA interference design,
such as the fonts twisted, distorted, extra lines, etc.. Before the experiment of
comparing the design of the before and after improvement, must be confirmed to meet the preattentive search features, that is to say the performance of the work does not be affected by the increase of interference.

This results showed improvement after improved text-based CAPTCHA will reduce the difficulty of human identification and effectively protect against cracking, and could reduce the reaction time and error rate as well, and the subjective preferences of the normal CAPTCHA to the good.

並列關鍵字

CAPTCHA ； feature integration theory (FIT) ； Optical Character Recognition (OCR) ； preattentive

參考文獻

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

Broadbent, D. E., & Broadbent, M. H. (1980). Priming and the passive/active model of word recognition, In R. Nikerson (Eds.), Attention and performance VIII.

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，

阮啟弘，(2005). 視覺注意力的研究議題與神經生理機制. 應用心理研究, (28), 25–50.

Chew, M., & Baird, H. S. (2003). Baffle Text: a Human Interactive Proof, 10th IS&T/SPIE Document Recognition, 305–316.