摘要
大部分的印表機使用半色調技術在紙張上表現灰階的效果，因此在論文中討論一些比較主流的半色調技術，並了解什麼原因讓這些技術會有視覺灰階的效果。 當我們掃描紙張上的資訊，儲存成電子檔，並且想要做影像壓縮、或一些圖像處理，我們會發現是不好處理的，因此我們會把半色調的影像轉回灰階影像，方便做之後的影像處理，該技術叫逆半色調技術。由於半色技術是一種失真的影像處理，因此我們不可能還原成原來的圖像，只希望盡可能地接近原來的圖片。論文中將討論五種逆半色調法，LMS、LMS-MMSE、LS、區域變異數分類訓練濾波器、高斯濾波器，並且比較它們之間的差異。前四種技術都需要透過樣本的統計，找出適合的逆半色調值，因此後續在論文中，我們根據邊界兩邊落差程度及邊界方向來做樣本分類，分類後的集合做出LS濾波器，達到更接近原圖的逆半色調圖片。
並列摘要
Most of printers use halftone technology to show the effects of gray scales on printed outputs. In this paper, we will discuss several main halftone techniques, and discuss why these technologies have visual effects of gray scales. When we scan the information on printed, documents and save them as files, we sometimes want to do image compression or other image processings. For this purpose, we need to convert halftone images back to grayscale images such that do image processings more conveniently. This technology is called inverse halftoning. Since halftone technology is a distortion image processing method, we can’t exactly original image, but to approximate as closely to original images. This thesis will discuss the five inverse halftoning methods: least mean square(LMS), LMS minimum mean square error(LMS-MMSE), least-squares (LS),the regional variance classification training filter and Gaussian filter methods. And we will compare the differences between those methods. For the first four methods we need the statistics of training samples to find appropriate inverse halftone values. Finally in the paper, we propose to do sample classifications based on the gaps of edges and edge directions. We then use the set after classifications to construct the LS filters whose resulting inverse halftoning images are closer to original images.