方格與蜂巢型影像細線方法研究

Abstract

細線化是圖形辨識中的一大課題，因其能有效的簡化資料並增強物體 特徵的萃取，尤其是在手寫文字辨識的應用上，本文中提出三種實行細線 化的方法：3x3 型罩法、矩形型罩法、蜂巢型罩法，並說明此三種方法的 處理原理、演算法與執行結果，再就這三種方法的優缺點加以比較分析， 歸納出其在應用方面最適切的使用方向；這三種方法的演算法型式皆相同 ，它們主要的不同點在於處理時所使用的判斷條件不同，其中 3x3 型罩 法與矩形型罩法在本論文中合稱為方格型罩法。由實驗結果得知，3x3 型 罩法對影像的處理最穩定，矩形型罩法處理後的影像點數最少，斷點亦增 多，蜂巢型罩法對端點的保留效果最好。此外，本文中也針對 3x3 型罩 細線化方法，提出一個改良式演算法，並可由此提速約 15%∼20%。 Skeletonization that compresses data and further supports feature extraction in the subsequent stage of image processing plays an important role of pattern recognition, especially on the application of OCR. In this thesis, three thinning algorithms are developed：each one for use with 3x3, rectangular, and cellular masks. In this thesis, 3x3 and rectangular masks thinning algorithms are refer to square thinning algorithm. The algorithmic approach to the development of each method is similar. Pictorial results produced by each of the algorithms are presented and their relative performances of these algorithms are also compared. It is found that the thinning operation by 3x3 mask is the most insensitive to image irregularities and noise, the thinning operation by rectangular mask results in the minimum number of pixels for patterns and the disconnection of patterns, and the thinning operation by cellular mask has the best effect of end point retention. Besides that, we proposed improved algorithm which could speed up about 15%∼20%.