隨著科技不斷的進步,對圖像處理的需求也隨之增加,而車牌辨識為圖像辨識常見的應用之一。辨識車牌號碼在生活上有許多應用,像是可以用於停車場收費、取締交通違規等等。因此提高車牌辨識的準確率可提升生活中的便利性、節省人力成本、提高執法效率。 本論文比較不同分類方法用於辨識車牌號碼的效果,期望能建立一套車牌辨識系統。因車牌圖像來自不同地區,則以資料增強方式增加訓練樣本以提高辨識正確率。車牌辨識可分為三個階段:偵測車牌區域、分割車牌字符、辨識車牌號碼。本篇分別探討各階段使用方法,分別以像素強度、方向梯度直方圖(HOG)做為特徵向量建立模型,再以指標評估哪一模型用於車牌辨識效果最佳,並探討各數字及字母的辨識情形。
As technology advances, the need of image recognition becomes important as well. License plate recognition is one of the common applications for image recognition. Identifying license plate numbers has many applications in life, such as parking fees, traffic violations, etc., thus improving the accuracy of license plate recognition can improve convenience in life, reduce capital expenditure, and improve law enforcement efficiency. This paper compares the effects of different classification methods for identifying license plate numbers, and hopes to establish a license plate recognition system. Since license plate images come from different regions, we increase the training samples by a data augmentation method to improve the recognition accuracy. License plate recognition can be divided into three stages: detecting the license plate area, character segmentation of license plate, and identifying the license plate number. This article explores the feasible methods at each stage respectively, using pixel intensity and histogram of oriented gradients (HOG) as the feature vector to build the models. We also use indicator to evaluate which model is best for license plate recognition and explore the identification of each number and letter.