文件影像中相連手寫數字之切割

Abstract

本論文提出了一個新的相連手寫數字之分割方法。首先，提出一個抽取數 字輪廓資訊的方法，建立上下輪廓線，取代數字的所有外輪廓點。此方法 可以將一張影像中最重要的資訊抽出，並且節省處理時所需之計算時間及 記憶體。接著介紹我們所提出的依筆劃方向變化切割相連數字之方法。從 已建立的上輪廓線中，尋找筆劃方向變化的像點作為起始的切割點，然後 我們利用垂直投影所得的像素統計量來決定切割點。此外，我們利用抽字 的方法來解決筆劃重疊的問題。接著，切割過後的數字區塊再經由我們所 提出來的方法將之結合抽取成單獨的數字區塊。最後，實驗結果的正確率 達96％以上。 A new approach to segmentation of touching handwritten numerals is proposed. First, a method for extracting numeral contour information is proposed.Upper and lower contour lines are created, which include most important numeralshape information with less demands of the processing time and storage.The point with the greatest stroke direction variation on the upper contour line is detected for use as the candidate segmentation point. Throughthe use of the pixel statistic obtained from vertical projection, the final segmentation point is decided. Furthermore, method for extracting numeral strokes from overlapping numerals and for extracting single numerals from multiple touchingnumerals are also proposed. Finally good experimentalresults achieving a segmentation rate over 96% are shown to prove the feasibility of the proposedapproach.