CAPTCHA introduction

近年來由於網路的蓬勃發展，提供了人們許多的便利，然而挑戰卻也隨之產生。有些人會利用“全自動機器人程式”（automated scripts/bots），大量的不當使用網路上的免費資源，使得伺服器的效能大大地降低，例如散佈垃圾郵件，張貼廣告等；也有人會藉由不斷的要求伺服器給予回應，進行“耗弱攻擊”（Denial of Service Attacks)來癱瘓伺服器。為了避免上述的不當使用，進而影響伺服器的運作，設計一套讓電腦能夠自動分辨回傳訊息是來自合理的使用者或是非正當使用的“自動機器人程式”的工具，就顯得越來越加重要。而“全自動公開化涂林測驗人機辨識”（Completely Automated Public Turing test to tell Computers and Humans Apart, CAPTCHA）就是在這樣的需求上發展出的產物。

The progress of the internet not only brings us more convenience, but also produces some problems. For example, the automated scripts/bots may abuse the free resource on the internet which will cause the surplus loads to the server, or someone may use the “Denial of Service Attacks” to paralyze server. To prevent these abuses, finding a way to help computer to identify the user is a legal human or an illegal bot program is become more and more important. The Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) is developed by this kind of application and needs.

Captcha主要的目的是透過電腦去分辨受測者是電腦還是人類，其典型的使用介面為一張印有文字的圖片，以及一個能夠輸入文字的編輯盒（TextBox）。圖1.1是目前雅虎奇摩（www.yahoo.com）在家族留言板上使用的Captcha系統，此系統會要求使用者回答在這張圖片中顯示了哪些文字。在此例中，為了擾亂電腦的辨識，將Captcha中的文字進行扭曲，同時加入一些雜線，因為目前的電腦在對這種文字的辨識率仍不高，於是機器人程式無法自動地正確回答系統所要求的問題；反之，因人類具有較佳的文字辨識能力，能正確地辨識這張圖片上的文字，在圖1.1的例子當中，一般使用者可正確回答，也就是在文字編輯盒內輸入BG8r6。所以，經由此機制，系統認為可以正確回答的便是人類，反之，一直答錯的就會被視為是電腦，將無法繼續接下來要完成的動作。

A typical user interface of a CAPTCHA includes a picture with characters and a text box. Fig. 1.1 shows a sample of the Yahoo’s [#] Captcha system. It asks the user to recognize the characters in the picture. In this CAPTCHA, some clutters are added and some characters are warped to interfere with the computer recognition program. Because the recognition rate for this kind of images is still low, the bots program can not easily answer this kind of CAPTCHA challenge automatically. In opposite, the human have better character recognition ability, so we can answer this question. No doubt, human can answer this question and input “BG8r6” in the text box. Accordingly, a CAPTCHA system identifies a request coming from a human if the user can answer the question correctly; otherwise, the system will classify it generated by a computer and the next operations will not be allowed. This system will ask the user to recognize the characters in the picture, and this example also warps the character and adds some clutters in the picture to interfere with the computer recognition problem. Because of the computer recognition rate for this kind of images is still low; the bots program can hardly to answer this kind of questions automatically. In opposite, the human have better character recognition ability, so we can answer this question more easily. In the fig. 1.1, human can answer the question and type the “BG8r6” in the text box, so the system will think the one who can answer the question is the human, otherwise, the one who usually gives wrong answers will be thought as a computer and it will not pass the test to the next process.