監視系統應用在路況監控已行之有年，但是因為通常路上的監視攝影機距離車輛很遠，再加上車輛移動造成的模糊以及其他雜訊影響，使得影片的品質不好，很難辨識出車牌上的文字。因此我們必須解決這個問題。 在這篇論文中，我們提出了一個不同的影像定位方法來應用在超解析度影像重建上。此方法是利用影像相關性比對以及比對之後的影像幾何轉換來使得所有低解析度影像都能重疊在同樣的影像座標上。定位完成之後，便將所有影像相加起來取平均，以得到重建的高解析度影像。最後我們利用一些後處理使得影像上的車牌文字可以更容易的被辨識出來。 我們拍攝了幾個測試影片來驗證我們的方法。拍攝方式是先在遠距離拍攝且故意讓攝影機失焦以得到模糊的低解析度影片。之後我們將實驗結果拿來跟雙立方內插之結果與之比較。最後實驗證明我們的方法可以成功的改善模糊的低解析度影像之品質，使得車牌上的文字可以清楚地辨識。 The surveillance video camera has been used on the road for traffic for a long time, but the resolution of the video is not fine enough to recognize the letters on the license plate. Addition to motion blur, noise effects and the distance of camera is far away from the vehicle, this becomes a very challenge problem. In this thesis, a novel approach for super resolution image reconstruction is proposed. We first register multiple blurred low resolution license plate images by correlation matching and geometric transformation with control point pairs, and then we fuse the registered low resolution images by interpolating and averaging them to high resolution image. After the high resolution image is restored, we apply the post process steps to further enhance the readability of the letters on the license plate. We capture low resolution test video sequence by defocusing the scene with vehicle to verify the proposed method. And we conduct a comparison to bicubic interpolation with the result by applying proposed method. The experimental result shows that our method can improve the quality of blurred low resolution image successively.
