標題: 彩色圖文美化剪輯與重排
Enhancement, Clipping, and Rearrangement of Color Document Images
作者: 黃善詩
Huang, Shan-Shih
蔡文祥
Wen-Hsiang Tsai
資訊科學與工程研究所
關鍵字: 文件了解;矩量保持;文章重排;彩色圖片品質改良;document understanding;moment preserving;article rearrangement;color picture quality improvement
公開日期: 1997
摘要: 本論文提出包含彩色圖片品質改良,文件了解和文章剪輯重排等功能的 系統.首先, 提出全彩階層式矩量保持的技術,可用以消除彩色圖片經過印 刷,掃描後的失真,並且自動地將重要顏色保留,達到減色的功效.若圖片的 顏色不多,再使用修改式k-means分群法,利用重要顏有群聚的現象,將最重 要的少數顏色保留,以達到壓縮的目的.並利用區域成長法,形態學的基本 運算等運作去除減色後可能出現的雜訊.在文件了解與重排方面,利用二值 化後水平與垂直的特性,將文件區塊抽取出來,再根據區塊的特徵辨識區塊 方向類別.並利用整體與部份投影,對標題與內文提出不同的切字處理.我 們也利用到中文文件版面設計的規則和區塊之間的關係,找出區塊排列的 順序.最後在重排的過程中,提供了數種格式,讓使用者可以選擇手動重排 或自動重排.藉由良好的實驗結果,我們證明了所提的方法是可行而且實用 的. In this study, a color document analysis system, including color picture quality improvement, document understanding, and article rearrangement isproposed. First, a full color hierarchical moment preserving technique is proposed to eliminate distortion caused by printing or scanning. Importantcolors can be preserved automatically and color reduction can be achieved.If the number of colors is small, a modified k-means clustering is proposedto preserve th most important colors, which is based on the observation thatimportant colors in common color documents tend to form clusters. Noise comefrom color reduction can be eliminated be the region growing method and someoperations of morphology. For document understanding and article rearrangement,we extract document blocks by vertical and horizontal projections and classifyblocks using some features of blocks. And then, global and local projections are used again to segment the lines and characters of blocks. Different processes are designed for segmenting characters in headlines and article contents. In finding the reading order of blocks, the knowledge of general layout rules, the composition techniques in Chinese documents, and the relationship of blocks are utilized. Finally, for the article rearrangementphase, users can choose serveral formats designed for manual or automatic rearrangement. Good experimental results prove the feasibility and practicability of the proposed approaches.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT860394097
http://hdl.handle.net/11536/62932
Appears in Collections:Thesis