流程圖的辨識與描述

Abstract

在本篇論文中，我們提出一套系統來辨認流程圖和表格的圖形部分。前處 理(preprocessing)包含了兩個步驟：輪廓追蹤(contour tracing)和線段 切割(line segmentation)。經過前處理後我們可以得到每個流程區塊( processing block)的線段組合。在第二個步驟， 我們抽取出這些線段組 合的特性。這些特性包含了流程區塊的內角和､組成這些流程區塊的線段 形式(line types)和這些線段組合的結合順序。我們提出一個方法來辨識 那些含有斷線的流程區塊。我們利用線段的方向碼(direction code)､各 種方向碼線段的長度和以及組成流程區塊的線段形式的順序來找出這類的 流程區塊。在辨識階段，我們分了三個步驟來辨識流程區塊､流程線段( flow lines)和箭頭。首先我們先利用區塊的內角和對圖形符號(graphic symbol)做分類。這些分類有平形四邊形流程區塊､非平形四邊形流程區 塊和非流程區塊。接下來我們利用這些圖形符號的線段形式組合來做辨識 。如果我無法在這個階段認出這個圖形符號，這些符號可能是有斷線的情 況。這時我們利用組合成這個符號的線段形式的順序來辨識。我們利用掃 描器來輸入影像。本系統對流程圖的辨率為79.8%(流程區塊的辨識率 為95.6%，流程線段的辨識率為84.9%，箭頭的辨識率為68.5%)，對表格的 辨識率為98.7%。 In this thesis, we present a system to recognize flow diagr- ams and tables. The preprocessing stage includes two steps: contour tracing and line segments extraction. Next, we extract the features of each processing block, flow line and arrow. From line segments, the feature include the sum of inner angles of a processing block, the types of lines which consist of the processing block, and the sequence of lines. We proprose a method to extract the processing blocks to find those processing blocks. In the recognition process, we use a three- stage matching algorithm to recognize processing blocks, flow lines and arrows. First, we use the sum of inner angle of a symbol to classify the symbol into one of the three categories of graphic symbols. The three categories are parallelogram processing blocks, non-parallelogram processing blocks and non- processing blocks. Second, we use the line types to recognize the processing blocks. Third, if we still cannot recognize the graphic symbol in the second step or those processing blocks which exist broken lines. We use the line sequence of processing blocks to recognize the graphic symbol. The recognition rate of flow diagrams is about for processing blocks, 84.9% for relation lines and 68.5% for arrows) and the recognition rate of tables is about 98.7%.