本帖最后由 狄霽顥嵚 于 2015-1-27 19:25 編輯
1、前言：因為在管理區看到投票驗證碼的帖子，想起來之前在技術討論區看到的一篇關于這個的文章，但是原帖找不到了，所以樓主自己搜集編輯了一下，大家看看漲漲姿勢吧~
常在網上飄，驗證碼相信大家幾乎每天都會見到，可是你知道么？當你每次不耐煩地輸入一個很難看清楚的單詞的時候，除了防止垃圾注冊或者評論以外，其實都在悄悄為人類文明做出一點貢獻呢。
驗證碼(CAPTCHA)，還有一個冗長的名字叫做“全自動區分計算機和人類的圖靈測試”(Completely Automated Public Turing test to tell Computers and Humans Apart)。在國外很多下載網站上，Google reCAPTCHA驗證碼使用得非常多，它可以免費申請，而悄悄做貢獻的就是它。@JimmyLiye編譯了Google的一篇文章，講述了這個乍一看很不可思議的小故事。
這事兒還得從OCR說起。OCR的意思是光學字符識別，簡單地說就是將圖片型的文字掃描并識別出來。問題是，這種技術現在的效率實在不敢恭維，經常錯誤百出。你看：
有一天，某臺機器掃描了一本書，想把它轉成電子版：
處理出來是這樣子的，勉強能看看：
The Hreckinridge' and Lane Democrats, having taken courage at the recent eastern advises, are [xxxxxxxxxx] energetically for the campaign: Several prominent Democrats who at first favoredDonoLea, are coming out. for the other aide, apparently under the [xxxxxxxx] of Federal [xxxxxxxxx]. An address to the National Democracy of ,1ifornia, urging the party to supportHaeeslipslDas, has recently been published, which manifestlybss strengthened that aide of the [xxxxxxxxx]: It is signed by 65 Democrats, many of whom occupy respectab e and prominent positions in the party, 22 of them are Federal office-holders,[xxxxx] more are recipients of Federal patronage, and the others represent a mass of politicians giving the document [xxxx][xxxxxx] mTheDcu8las Democrats are also active The Irish and German vote will mostly go with ths# branch of the party, but it is[xxxxxxxxx] to [xxxxxxxx] [xxxxx] [xxxx] [xx] the stronger. Thus far 17 IT newspapers have declared for DonGres, 13 for Base$- IaaIDGS and 9 remain non-committal, with even chances of going either way. Under these circumstances the Republicans entertain not unjustifiable hopes that the Democratic divisions may be so equal,- ly balanced as to give the State [xx] LIaCOLV.Same very [xxxxxxx] Bell and Everett meetings have been held in different parts of the State, bat thus far that party does not exhibit much rank sad ale air en.
還有這個，原書質量比較差：
看到這個，電腦就傻眼了，吐出來一堆亂七八糟不知所云的東西：
‘ letz-1- rrk fit: 1′ . on its to Vc ,rt, cann into tlm yc H_ tcr,la, .n. ‘l l; , arc ti:( h of thc 1″,ats that to ltc rc: ,;. , I; ., l: rel!;n. tani., , ./olio, IJuteilu, . 1!’i./_ ;lr”n. Iiam! Jr.r. F’l,nr_.Z.._%i;;, ,, : rt-Irn: am/ tf.rri.:, t?m steamer as a tr nW r. Uu ,tin;t, c ac?1 1″,at firm/ a t;nn, accor.liu; to .t rn. ‘Cl.w r. wu ru lm:nui MistinW /y in u;th, -. ink ;:,k as to “what w ax 1111, :111(I vle:iR a of ;: (,am( into, mnr r-, tm if tlm wo r( uu.i n:’ of t?u : la?:Iv. \ ‘c : ol in thc , ucr:atic , , Tlau :; will h:aw tu-li.r \. ’1′Im yap?tts Il ,,n an,/ I, ,rr:l. r, (,t tf,is r:ity, start witli it, with lu:rtic: ol \ 1- e:l.k.
看得懂嗎？反正我是看不懂。
reCAPTCHA驗證碼的目的之一就是為了改變這種情況的。下邊這張圖可以很好的解釋它的工作原理：1、掃描書籍；2、提取OCR無法識別的單詞；3、進一步扭曲并加入隨機橫線來增強安全性；4、使用兩個單詞生成驗證碼來讓用戶識別。
有了它的幫助，第二張圖片上面的文字就基本順暢了(盡管還是有一點小錯誤)：
The New-York State yacht Squadron, on its annual cruise to Newport came into the harbor yesterday afternoon. The following are the names of the boats that came to anchor here: Jessie, gera loliv erelun Annie, Mannering, Julia, Bonita, Magic wut, Rambler, floumblie, Henrietta, Sea-Drift and Maria, with the steamer America as a tender. On anchoring each boat fired a gun, according to custom. The reports were heard distinctly in the city, causing considerable inquiry as to "what was up", and quite a number of sanguine individuals came into our office to inquire if the guns were not annunciatory signals of the successful laying of the Atlantic Cable. We invariably replied in the negative. The squadron will leave to-day for Newport. The yachts Washington and buub r of this city, start with it, with parties of New Haven people.
有的人可能要問了，既然機器都看不明白，那他怎么判斷你輸對了還是錯了呢？這個問題問得好，Google的解決方法也很絕：
兩個驗證碼里面有一個是正確的，被人審核過的，而另一個是不正確的，機器讀不出來的。當你把那個正確的輸對以后，我們就會默認另外一個也是對的。這樣，你每輸入一次驗證碼，就為人類的知識寶庫增加了一個單詞。
（不是一個用戶識別一個單詞，是很多個用戶識別，就算每個人認為的都不同，但數量越多最后的結果就越接近正確答案。。。——樓主注）
2.第二段的大意就是驗證碼可以不用取消，改改方式。（搜資料時偶然看到想起的）
百度輸驗證碼的時候就是這種方式，用漢字，下面是個九宮格，直接點擊匹配的漢字就ok*先裝個sp1，一會編輯* |
好高深~完全不在一個level~看不懂~
