違規停車照片之車牌辨識

Abstract

儘管車牌辨識系統已有多年的研究，但對於背景複雜和不限方向、位置和大小的車牌之影像分析以及低解析度車牌字元的辨識，卻較少有研究和討論。在這篇論文中，我們提出一個違規停車照片之車牌辨識技術。本論文主要包括車牌定位、字元擷取之前處理及字元辨識三部分之研究。在車牌定位的部分，我們先利用了雙差分濾波器找出車牌的位置。在字元擷取之前處理的部分，我們利用差分濾波器來找出每一個前景線段的起點和終點，以達到分割前景和背景的功能。最後，在車牌字元的辨識上，對車牌切割所得的車牌字元影像，我們應用列處理法來擷取其特徵，然後再以樹狀辨識器來作辨識。本論文中所使用的測試樣本為員警所拍攝的違規停車照片，對於各式車牌及各種背景環境，大致算得上有不錯的辨識效果。

In this paper, the approach to automatic license number recognition for car images taken for parking violation is presented. The approach consists of three phases. First, we use a dual difference filter to enhance the license plate in the car images for the license plate location. Second, for the segmentation of the license plate image into foreground and background, we use difference filter to find the two end points of each horizontal segment of the foreground. Finally, for character images segmented from the license plate, a row-based method is used to extract the feature of the character, which is then applied to a tree classifier for recognition. For real car images of different types of license plates and various background scenery, the proposed approach seems to yield satisfactory results for license plate location and recognition.