基於影像切割結果對文件影像內容作分析

Abstract

本論文提出一個文件分析系統，包含自動校正文件影像偏斜、文件切割、分類、了解與展示。首先我們提出一個利用赫佛轉換 (Hough transform) 偵測出文件影像偏斜角度並予以修正的方法。在文件切割過程，我們利用一個由下而上對彩色文件做切割的方法，取得輸入文件影像的文字區塊、文字列、圖類區塊等三種基本區塊。接著在文件分類過程，我們利用各種不同的特性將文件中的標題、表格與小塊文章從上面提到的三種基本區塊中抽取出來，然後利用一個文字辨認系統辨識標題中的文字。有了分類區塊與標題文字以後，使用者可以透過我們所提供的操作界面，將原來的文件以文字方式呈現標題，而以分類過的區塊呈現其他部分。此外，我們還提出一個保留圖形邊緣資訊與使用文字呈現標題的縮圖製作方法，來提高縮圖的視覺效果。由實驗結果證明本論文所提出的方法是可行且實用的。

In this study, a system for document analysis, including skew correction, segmentation, classification, understanding, and display is proposed. In the skew correction phase, we propose a data reduction method for fast skew estimation using the Hough transform. In the segmentation phase, a bottom-up method for color document segmentation is adopted to obtain segmented blocks, including text blocks, text lines, and graphic blocks, of the document image. And then in the classification phase, several features are used for extracting titles, tables, and small enclosed articles from segmented blocks. After block classification, titles are understood by an adopted OCR system, and with a user interfaces designed in this study, the document can be displayed conveniently with classified blocks. In the thumbnail creation phase, we propose a novel method to create a thumbnail image with better visual effects by keeping edge information in graphics and table blocks, and showing ASCII characters in titles. Experimental results are shown to prove the feasibility of the proposed approach.