摘要
半色調技術是種能夠將連續色調圖片轉成只有黑跟白兩種色調的二值化圖片的技術，本論文將於第二章介紹。而逆半色調技術則是反過來將半色調圖片重建成連續色調圖片，由於是藉由模擬樣本灰階圖片的灰階值做重建動作，所以它是一種失真的影像處理。第三章將會介紹一些重要的逆半色調技術如LMS與LS。接著將介紹基於圖片資料特性去分類還原更好的逆半色調圖片的技術，本論文將優先介紹近期較知名的基於資料分類之逆半色調技術Guo的變異數分類法。 接著本論文第四章介紹基於邊界資料分類的逆半色調技術，包括Sobel及Laplacian，這種兩種方法可以對高頻及低頻區域做較為正確的還原動作。Ong提出了Sobel邊界分類法並訓練出397個濾波器，但是，397個濾波器顯然過多，訓練時間也會增加。本論文將基於角度跟梯度這兩個參數，探討簡化分類數目的可能性，會簡化其中一項分類數量來檢視哪種參數影響分類的效果較為顯著，最後本提出的Laplacian邊界分類法跟Sobel邊界分類法做比較，這些基於資料分類的逆半色調技術實驗結果將呈現在論文結尾並作深入探討。
並列摘要
Halftoning technology is a method to convert gray-level images to binary images. Well-known haftoning technologies will be introduced in Chpater 2. Inverse halftoning is a technology to reconstruct continuous tone images from halftone images. Because it is a method based on pattern-training to estimate gray-level values, it is classified as am distorted image. In Chapter 3, we introduce recent inverse halftoning technologies based on pattern training including LMS and LS. Then we will discuss inverse halftoning based on data classification. First we will introduce a recent well-known method: variance classification by Guo. In Chapter 4, we discusss inverse halftoning methods based on Sobel operator and Laplacian edge data classifications. In Ong’s paper, 397 filters are trained by Sobel edge classification. In this thesis we will discuss methods to reduce filter numbers, and compare to Laplacian edge classification. Experiment results of inverse halftoning based on data classifications will be discussed and showed in the end.