書法影像內中文字的二值化和非等方向性擴散

Abstract

在這篇論文裡我們設計了一種方法來對書法影像做二值化，二值化之後的結果可以被用在其他的應用程式上：例如邊緣描述和影像壓縮。首先我們會分析書法影像內的屬性，然後偵測出哪一種二值化的方式適合使用在書法影像中。書法影像被歸類為兩種類型：碑文和字帖，前者是黑底白字而後者則是白底黑字，因為書法影像中有許許多多不同的雜訊和字元間灰階度差異相當的大，所以在做完二值化之後，所得到的前景影像將會是破破碎碎的，我們將會使用非等方向性擴散的技巧來模糊原始的書法影像，這樣將可以把字體邊緣保留住而不被模糊處理所影響，最後我們在做完二值化之後，利用字元寬度的資訊來去除雜訊。在我們的系統實作測試了90張碑文書法影像和51張字帖書法影像，實驗結果顯示這個系統相當的有效能夠得到很滿足的二值化結果。

In this thesis, we propose methods for Calligraphy document binarization. The results can be used for other applications such as shape description, and image compression. First, we analyze the properties of calligraphy documents, and then determine the binarization technique to be used in specific images. Calligraphy documents are classified into two classes: tablet documents and writing documents. The former have black background and white foreground, and the latter have white background and black foreground. Because calligraphy documents have various noise and large variations in gray scales of the characters, the foreground will be broken after thresholding. We use the anisotropic diffusion technique to smooth original calligraphy documents which can preserve the boundary of characters well. Last, we remove noises from binarized images based on the information of stroke widths. In our experiments to 90 tablet documents and 51 writing documents, we can obtain satisfactory binarized results of calligraphy documents by using our procedure.