以筆畫線段為基礎的手寫中文字旋轉偵測

Abstract

隨著平板電腦、手持式行動裝置的日漸普及，手寫文字辨識的需求也日益增加，這樣的需求在使用漢字的地區更加迫切。當漢字手寫識別技術的應用日趨成熟時，建構一個自由自在的書寫輸入環境的需求也就慢慢浮現出來。本研究的目的，是希望在進行手寫識別之前，將輸入軌跡的旋轉角度找出來加以校正，讓使用者不需要把輸入裝置面向自身的方向來書寫，也能夠得到正確的識別結果。本研究中使用漢字筆畫線段中的基本特性，由左至右的水平筆畫以及由上至下的垂直筆畫，它們的垂直關係是不會隨著軌跡的旋轉而變化的，只要確立了這兩種筆畫目前的角度，旋轉角度也就可以偵測出來。本論文的實驗，首先要找出輸入軌跡的筆畫線段的方向與長度，其次使用兩種不同的方法，筆畫線段投票法搭配經驗法則與K-mean clustering的統計方法，找出水平與垂直筆畫目前的角度，比較兩者得到的結果的優劣以及在處理速度上的差異。另外引入兩個輔助的實驗，其一是驗證手寫識別程序傾斜軌跡處理效果，以證明本實驗的方法是真正有效的；另一則是針對輸入軌跡使用PCA方法嘗試找出旋轉角度做為本論文方法的對比。最後，從兩個偵測方法中找出一個較佳的方法來建構一個手寫中文字旋轉偵測系統。

As TabletPCs and Handheld devices are widely sepreaded, handwriting characters recognition systems are requested,and the request is more important in areas using Chinese Characters.More and more stablized applications of Handwriting recognition technology makes the requirement.A more free handwriting input environment is prompted.This research focuses on rotation angle detection during the preprocess of handwriting characters. Users need not turn the input device to themselves before starting input.Based on the basic characteristics of input strokes,segments of Chinese characters, horizontal strokes from left to right and vertical strokes from top to bottom,this research can detect the rotation angle if such invariant relations are found.First of all is to compute the direction and length of each stroke segment of the input writing.Secondly, two different methods and their combination, structure analysis with heuristic rules and K-mean clustering, are used to detect the rotation angle. Two associated sets of experiments are performed.One is verifing skew/rotation limitation of the handwriting recognition engine to make sure the research is working. And the other is appling Principal Component Analysis on input writing to detect the rotation angle as contrast.Finally, comparing these two results and pick the better one to construct a Handwriting Chinese Characters Rotation Detection System.