違規停車JPEG影像之車牌分析

Abstract

儘管車牌辨識的技術已經發展多年，但是對於解析度偏低，且沒有固定環境背景的車牌影像的處理，卻仍是一個較少人研究和討論的方向。在這篇論文中，我們發展一個車牌自動分析系統，針對JPEG格式的違規停車壓縮影像，做車牌定位以及車牌字元萃取的處理。其中，我們利用影像解壓縮的過程所獲取的資訊，直接在壓縮域判斷車牌的可能位置，並只針對這些區域完整解壓縮，以減少大量的計算。另外，我們增加字元影像的平滑性，以解決低解析度影像所造成切割出來的字元過於粗糙的問題。依據我們實驗的結果顯示，本篇論文所提出的系統，對於車牌自動分析的效果相當好。

In this paper, we develop an automatic license plate analysis system for JPEG compressed images taken for parking violation. The proposed method is focused on the parts of license plate location and character extraction. A robust method for the segmentation of the license plate in the compressed domain is presented to reduce the expensive operations of inverse DCT. To overcome the problem of poor quality of the extracted characters caused by the low resolution of each character, a smoothing process is developed. According to the experiments carried out for some real images, the developed system is quite effective for license plates analysis.