車牌辨識系統上字元切割、導正及辨識之研究

Abstract

由於車輛持續的增加，如何利用自動化的車牌辨識系統來管理車輛是一種趨勢。而車牌辨識系統的研究已有多年，但仍有許多缺點待改進。在本論文中我們發展出一套硬、軟體更新便利且移植性高的車牌辨識系統。 本論文是針對車牌辨識系統中的水平軸導正處理、垂直軸字元切割及字元辨識三部分做研究。水平軸導正處理之目的是將車牌字元上、下緣以外部分切除，並將字元導正到相同水平軸上，方便後面的垂直軸字元切割處理。在垂直軸字元切割處理上我們提出了模版比對切割及區域生長、分群切割兩種方式，各有適用範圍。在本車牌辨識系統中最後是以正確性較高的區域生長、分群做切割處理。因為車牌字元的字型及大小幾乎固定，因此採用比對的方式並輔以外型及其他特徵做字元辨識，效果不差。 在本論文中所使用的測試樣本高達362張車牌影像，都是由實際車牌辨識管理系統拍攝所得，並包含各種不同環境及車牌種類，因此本實驗結果之可靠度及可行性都很高。

On the increase of car, so how to use automatic license plate recognition system for car's conservation is important. Although the history of license plate recognition system research is long, the system still has many faults. In our study, we develop a license plate recognition system whose hardware and software update conveniently. The research of thesis focuses on Character Partition Horizontal Normalization and Recognition. The purpose of Horizontal Normalization is to cut the region that is out of the character horizontal edge. And let the horizontal of character are the same. After this process the character partition and recognition is simple. On the character partition research, we use two methods. One is using template map to partition the character of license plate. The other is using growing region method to partition the character. Both of those two methods have advantage and disadvantage. The result of our system is using growing region to partition the character of license plate. Because of this method has batter adaptability and correct percentage. Recognition is the last part in our process. Because of the size and shape of character is almost the same, so we use statistics to create the stander map character, and then recognition these character with map. But this method does not recognize the shape similar group like O. D. Q and 0. So we add shape and other particularity to redouble recognition. In our study the testing samples include 362 pieces of license plate image those were come form realistic parking lot's monitorial system, wherefore our experimental inference is dependable and our implement system is realizable.