利用投影正交相關法之邊緣導向傾斜矯正技術

Abstract

現今每日收送的辦公文件為數相當龐大。由於這些文件以紙上形式為多，故發展文件影像分析系統，將紙上形式的文件轉換成電子形式，其重要性急遽的增加。在文件掃瞄的過程中，我們必須確保文件方向的正確性，以避免後續文件分析上的錯誤發生。本論文提出了兩種偵測二值化文件中之傾斜角度的方法。第一個方法是利用模糊c迴歸模型 (fuzzy c-regression models) 來偵測傾斜之角度。第二個方法是依據投影之字行間正交相關 (projection-based interline cross-correlation) 法於掃瞄之影像上。此法針對隨機選取之小區域做相關性的運算，一改從前針對整張影像作運算的缺失。所提出之兩種傾斜角度偵測方法皆不需要圖文分離之繁複的前處理步驟，可大幅地降低其複雜度。於本論文中，將介紹一些文件分析技巧之前處理步驟，包括 run-length smoothing 、 black-white transition 、 及一些區域選擇之方法，這些技術將使得所提出之傾斜矯正系統所適用的文件型態更加的廣泛。最後，一新的內差技術，稱為以邊緣為依據之比例雙線性內差 (edge-directed ratio bilinear interpolation) 技術，將於本論文中詳細的描述。

Since the number of daily-received paper-based office documents is overwhelming, the development of document image analysis, which converts the paper-based documents into electronic forms becomes increasingly important. During the scanning process, we must assure that the document is in the right orientation to avoid mistakes in the following analysis. We describe two algorithms for skew detection in binary document images. The first method is based on fuzzy c-regression models (FCRM). The second method is based on projection-based interline cross-correlation in the scanned image. Instead of finding the correlation for the entire images, it is calculated over small regions selected randomly. For both two methods, they do not require prior segmentation of the document into text and graphics regions and greatly reduce the complexity of the operation. In this thesis, several image analysis techniques, including run-length smoothing, black-white transition, and operating window selection, are used to broaden the document types for skew detection. A new interpolation technique, which is called edge-directed ratio bilinear interpolation technique, is also presented here.