透過您的圖書館登入
IP:54.211.203.45
  • 學位論文

以不完整的內容檢索資料庫內的車牌號碼

Retrieval of Vehicle License Number from a Database Using Imperfect Input

指導教授 : 陳世旺
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


視覺式的車牌辨識是一件看似簡單的工作,但是要達到高辨識率非常困難。視覺式的系統受到光線變化的影響很大,百分之百完美的車牌辨識系統並不可得。然而,在某些特定的應用中,當車牌辨識的工作能與車牌資料庫搭配時,即使不完美的辨識結果仍有很大的用處。舉例來說,假設在視覺式的智慧型停車場內所有停放的車輛,其車輛資訊皆已由視覺式的系統記錄到資料庫內,則車主取車時,我們可以依據車牌號碼之間的編輯距離很快地從資料庫中找出與輸入車牌最相似的候選車輛。 編輯距離是一種有效測量字串之間相似程度的工具,當比較車牌時,我們採用Chamfer distance來定義字串編輯時所需花費的「插入字元」、「刪除字元」與「取代字元」的代價。因為Chamfer distance能反映兩張影像在形狀上的差異,因此兩張車牌號碼的編輯距離能代表了兩張車牌號碼在形狀上的相似程度。 在本文中,我們改善編輯距離的計算方式,將字元與其鄰近字元的關係納入考慮。此種計算編輯距離方法最早由J. Wei[Wei04]提出,稱之為馬可夫編輯距離。我們修改了J. Wei的論文兩個有關馬可夫編輯距離的派系能量函數,使得修改後馬可夫編輯距離適用於車牌號碼的比對,而且其結果也較傳統的編輯距離的值更加精細。此種馬可夫編輯距離能有效地反應出車牌號碼之間的號碼錯置的關係。

並列摘要


Visual-based car license plate recognition seems to be an easy work, but practically speaking, it would be a hard work to have high recognition rate. Luminance condition has a huge influence over all visual-based systems, such that a perfect car license plate recognition system is still unavailable. In some specific applications, these unstable results could be useful if the work of recognition was supported by a license plate database. For example, a database can contain vehicle information of all cars parking in visual-based intelligent parking lot, when a vehicle owner wants to take his/her car, we can retrieve candidate cars from database according to edit distance between input license plate and license plates in database. Edit distance is a powerful tool for measuring the similarity between two strings. When comparing two license plates, we use edit distance technique to calculate the difference of two license plates using Chamfer distance to define the character editing cost which contains 「inserting character」、「deleting character」 and 「replacing character」. Then the edit distance of two license plates can represent the similarity of them. We modify the method of calculating edit distance by considering the neighborhood relationship of characters in source string when editing source string to destination string. The proposed method is first called Markov edit distance by J. Wei[Wei04]. We modify two clique potential functions from J. Wei’s paper to fit license plate comparison and get the finer edit distance when Markov relationship is considered. The modified Markov edit distance is very useful when compared license plates with reshuffling numbers.

參考文獻


[莊01] 莊建宏, 自動化交通監控系統 (Automatic Traffic Monitoring System), 國立臺灣師範大學資訊教育研究所碩士論文, 中華民國九十年六月 (2001)
[Mye00] Richard Myers, Richard C. Wilson, and Edwin R. Hancock, Bayesian Graph Edit Distance, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 22, No. 6, 2000, 628-635
[Cha91] S. C. Chan, A. K. C. Wong, Synthesis and Recognition of Sequences, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 13, No. 12, December 1991, 1245-1255
[Che98] S.W. Chen, S. T. Tung, and C. Y. Fang, Extended Attributed String Matching for Shape Recognition, Computer Vision and Image Understanding, Vol. 70, No. 1, April 1998, 36-50
[Gar03] Jose Romon Garitagoitia, Jose R. Gonzalez de Mendivil, Javier Echanobe, Jose Javier Astrain, and Federico Farina, Deformed Fuzzy Automata for Correcting Imperfect Strings of Fuzzy Symbols, IEEE Transactions on Fuzzy Systems, Vol. 11, No. 3, 2003, 299-310

延伸閱讀