利用熵選取門檻之自動彩色邊緣偵測

Abstract

本論文，我們提出基於向量階層統計與主要成分分析的彩色邊緣偵測技術，並且利用熵達到自動選取門檻的自動彩色邊緣偵測。利用我們提出的自動彩色邊緣偵測，不僅可以偵測到當相鄰物體具有相近的灰階值但不同色彩的邊緣，且門檻是依據影像內容所自動最佳調整，而不需要手動選取，增加使用者的方便性與信賴度。

In this thesis, we propose automatic color edge detection techniques based on vector order statistics and principal component analysis by entropic thresholding. Both methods employed improved entropic thresholding to determine the edge threshold. Our color edge detection techniques can detect edges when neighboring objects have different hues but with similar intensities, which cannot be detected by known grayscale or color edge detectors. Furthermore, by using entropic thresholding we can automatically determine an optimal threshold which is adaptive to different image contents without manual intervention. Edge detection by our proposed scheme is very user friendly and confident.