高對稱對比區域對稱偵測之加速處理

Abstract

典型的對稱偵測程序大致上分為三個階段：(1)特徵點篩選(高對稱對比區域)(2)由特徵點進行對稱資訊計算(3)決定對稱所在及對稱範圍。特徵點之多寡決定了整個偵測程序之時程，而通常將特徵點數目不在少數，因此成為瓶頸所在。 本論文研究課題旨在篩選特徵點，有效減少特徵點數量，同時不至於嚴重影像對稱偵測。本論文定義極值點，並結合亮度對比與特定梯度結構分佈，有效將特徵點大幅降低，並達成與實驗室既有對稱偵測系統匹配之準確度。

The speed of a symmetry detection process is in general determined by the number of feature points(Potentially Symmetrical Areas), on which symmetrical information are estimated for determining the existence and location of one (or more) axis of symmetry. The study of feature point reduction by the proposing the use of local max/min points in association with contrast and a specific pattern in gradient structure will be reported in this thesis. The amount of reduction in feature points was found to be significant as compared to an existent symmetry detection system in the lab without degrading the performance in the accuracy of symmetry axis computation.