臺灣博碩士論文加值系統

在這篇論文中, 我們提出一個以具有週期性特性的色調(hue) 屬性為基礎
的彩色影像分割方法. 首先, 我們根據均勻色度的知覺色彩空間 (UCS
perceptual color space)建構出一個色調的循環式直方圖(hue
circular histogram). 然後,用比例空間濾波器(scale-space filter)來
自動地平滑(smooth)直方圖, 並且將此循環式直方圖展開成一傳統的直方
圖. 最後, 再根據變異數分析法(analysis of variance)的判定準則重複
地定階直方圖. 在論文中我們做了三組實驗: (1) 我們提出來的定階方法
在循環式直方圖與傳統直方圖上定階之比較; (2) 循環式直方圖定階法與
群聚分類法(clustering)效益之比較; (3) 採用均勻色差與非均勻色差色
彩空間的比較. 從實驗中可以證實我們所提出的理論正確 , 而我們所提
出的循環式直方圖定階法在分割彩色影像時具有相當不錯的效益.

A circular histogram thresholding for color image segmentation
is proposed. At first, a hue circular histogram is constructed
based on an UCS (I,H,S) color space. Next, the histogram is
smoothed by a scale-space filter, then the circular histogram
is transformed to a traditional histogram form. At last the
histogram is recursively threshold based on the maximum
principle of analysis of variance. Three comparisons of
performance are reported, which are (1) the proposed
thresholding on the circular histogram and a traditional
histogram; (2) the proposed thresholding and clustering; (3)
thresholding on hue attributes of UCS and non-UCS color
spaces. Several benefits of proposed approach are identified
in the experiments.