彩色雜誌影像之分割與了解

Abstract

本論文提出了彩色雜誌之圖文分割與版面分析的方法。首先，提出一 動態分割演 算法，可將彩色雜誌影像內容作更合理的分群，然後利用新 提出的區塊分割演算法將 文件中相同屬性且鄰近的像素更有效率的連接 成許多區塊，並依這些區塊的結構與統 計特徵判斷出其為其為文字或圖 像區塊，再根據雜誌較多複雜背景的特性，將同屬性 且鄰近區塊合併成 大區塊。接著利用這些區塊來分析版面配置。首先依據大部份文字 的走 向求知文件列印方向，再利用區塊文字的大小和區塊的位置等特徵找出文 章標題 、摘要、章節、抬頭、頁腳、頁碼及圖的說明等。在處理多頁時 亦提出將跨頁和跨頁 圖連接起來的方法。藉由實驗結果，可以證明所提 出的方法是可行的。 An intergrated approach to segmentation and understanding of color magazines is proposed. First, a dynamic-cut color quantization method is proposed, which can be used to classify color image contents into more reasonable classes. For block segmentation, an new algorithm is proposed to grow neighboring pixels with indentical colors into blocks more efficiently. After block extraction, several features are used to classify extracted blocks into text blocks and graphic blocks. For construction of blocks magazine images which usually have complicated background, improved methods for merging textblocks into text areas and graphic blocks into graphic areasare also proposed.These classified areas are then used to analyze the document layout. The printorientation of the document is detected according to the print orientation ofmost textline blocks. Methods for finding titles, abstracts, section titles, section texts, headers, footers, page numbers, and captions of graphics inmagazines are proposed. For multiple-page documents, methods for linking textcontents and graphics which are in adjacent pages are also proposed.Experimental results are shown to prove the feasibility of the proposedapproach.