彩色文件影像分析所需圖文及特別特徵之分割

Abstract

本論文提出了一個彩色文件影像分析所需圖文及特別特徵之分割系統。首 先，提出一改良之中間值分割演算法，此方法可用以將彩色文件影像作分 群。接著介紹我們所提出的圖相接、調整傾斜文件影像及偵測文件影像列 印方向之方法。然後我們利用多頻譜有限制長演算法來抽取彩色文件影像 中之物件，並且利用一些統計特徵及一個結構特徵來辨識文字區塊與圖形 區塊。辨識過後的文字區塊再經由我們所提出來的方法將之結合成完整的 文字區塊。最後，本論文提出了用以辨識文字區塊型別與偵測文字區塊列 印方向的方法。實驗結果證實所提技巧確實可行且具實用性。 A system for segmentation of texts, graphics,and special components in color document images is proposed. First, a modified center-cut color quantization method is proposed, which can be used classify color image contents into several classes. Then proposed methods for combining subimages into complete images, adjusting skewed document images, and detec ting print orientations of document images are described. For block segmentation, a multi-spectral constrained run length algorithm is used to extract objects from multiple color planes. After block extraction, several statistical features and a structural feature are used to classify text blocks and graphic blocks. Then methods for merging text blocks into text areas are proposed. Finally, methods for detecting the print orientations of text areas and for classifying their types are proposed. Some experimental results are shown to prove the feasibility and practica- bility of the proposed approach.