病歷表影像的分割與內容元件的分類

Abstract

本論文提出了病歷表影像分割與其內容元件分類的方法。首先，提出 一修改的赫夫演算法，可以偵測出病歷表影像中有文字雜訊干擾的連續線 段。基於這個演算法，我們又提出了抽取影像中表格框線的遞迴演算法， 可以結構化地抽取表格框線，並以樹狀結構排列。在抽取表格框線後，我 們提出了幾種修補遮罩來修補表格框線抽取過程中所造成的破碎字，並以 尋找連接元件的方式，與幾項合併技巧，將影像中的內容元件分割出來。 最後，我們提出幾個用來分辨預先印刷元件或是手寫填入元件的特徵，與 一個倒傳遞類神經網路，將病歷表影像中的內容元件分類。藉由實驗的結 果，可以證明我們所提出的方法是可行的。 An integrated approach to form segmentation and component classification forclinic data image analysis is proposed. First, a modified Hough algorithm isproposed, which can be used to detect consecutive line segments from noisy formsin a clinic data image. Based on this algorithm, a top-down recursive form frameextraction algorithm is proposed, by which form frames in a clinic data imageare extracted structurally. After form frame extraction, a restoration maskingmethod is proposed, which is used to restore the broken strokes resulting fromframe extraction. Then, a component extraction and merge method is proposed forform content extraction. Finally, several features and a back propagation neuralnetwork are used for classification of preprinted components and filled-in hand-written components. Experimental results are shown to prove the feasibility ofthe proposed approach.