從灰階文件影像中抽取二值化影像

Abstract

文件至今仍是最主要的媒體，而許多文件影像的應用諸如文字辨識及一些文件分析演算法只接受二值化影像，本論文提出一個從灰階文件影像抽取出二值化影像的方法。我們先將輸入影像做平滑處理再用影像的梯度值求出三階影像（Three-level image），之後再利用筆畫在小區域中的寬度應該相似的特點偵測出各方向的筆畫以及筆畫交叉點。由實驗顯示，我們提出的方法在許多受到污染的文件影像有不錯的結果。

Documents have been the most popular information median in our society today. Numerous applications of document image, such as OCR (Optical Character Recognition) and many document analysis algorithms are base on binary image. This thesis provides a method to extract binary images from grayscale document images. First, we smooth the input image and generate a three-level image using the gradient, then we use the property that stoke width in a small region is similar to detect all stroke and cross pixels. According to the experiment results, the proposed method gives good results in many polluted document images.