影像主成分色彩分析及其色彩意象之關聯性研究－以景觀建築影像為例

近期有關色彩意象研究逐漸採用彩色影像來檢視傳統以單色樣本建立的色彩意象尺度，並針對雙色或多色影像樣本進行色彩意象與色彩成分分析，同時透過意象尺度建立影像分類檢索，以及將色彩意象對映到CIE色彩空間的色度呈現。本研究擬針對特定彩色影像類型（景觀建築影像）檢視色彩意象尺度，透過受測者的色彩意象判定數據歸納分析色彩意象構面因素，同時運用像素分色法在三種色彩空間中分別找出影像樣本主成分色彩，並探討影像主成分色彩與色彩意象之關係。本研究收集112個景觀建築彩色影像為樣本，先進行一系列色彩心理物理判定實驗，獲得165,648次景觀建築意象判定數值，再運用因素分析法推導出以「光亮－黯淡」、「東方－西方」與「美的－醜的」三維度為主要色彩意象構面因素，同時印證與Osgood語意空間（Semantic Space）之三屬性具有一致性。經過交叉分析後，顯示本實驗色彩意象三維度之間具有良好的相關性。

關鍵字

色彩意象；色彩心理物理實驗；主成分色彩分析；色彩空間；景觀建築影像

並列摘要

Recently, color imagery research has gradually used complex color images to examine the color imagery scale which traditionally deducted by monochrome patch samples. Researchers tend to analyze color imagery factors and principle component color of two-color or multi-color image samples. Meanwhile, researchers apply the color imagery scale to constructing the image classification index and calculating the colorimetric values in the CIE color space. The current study intends to examine the color imagery scale for a specific color image type (landscape architecture image) and to derive the new color imagery factors from the judgment data of test images. The use of the pixel-color separation method helps to analyze the principle component color of tested images in each of color spaces. The further study is to investigate the correlation between the principle component color and the color imagery of tested images. There are 112 landscape architecture color images collected as test samples. First, a series of color psychophysical judgment experiments carried out. A total number of 165,648 judgment values of tested images were gathered. The study adopted the factor analysis method to derive a three-dimension of the color imagery factors as ＂Light-Dark＂, ＂Eastern-Western＂ and ＂beauty-ugly＂. It is consistent with the three attributes of Osgood Semantic Space proposed. After a cross-analysis, the results obtained the good correlations among the three-dimension of the color imagery factors in this experiment. Further, the principle component color of landscape architecture images was analyzed in CIE L*a*b*, CIE L*u*v* and YIQ color spaces by pixel-color separation software. The relationships among color attributes are as follows: (1) The ＂light-dark＂ dimension has a good correlation with the lightness and chroma in both CIE L*a*b* and CIE L*u*v* color spaces, but hue is in a lower correlation. The ＂light-dark＂ dimension has a moderate correlation with the chroma in YIQ color space, but a lower correlation with the lightness and hue. (2) The ＂Eastern-Western＂ dimension is moderately correlated with the chroma in each of CIE L*a*b*, CIE L*u*v* and YIQ color spaces, but is lowly correlated with the lightness and hue. (3) The “beauty-ugly” dimension is moderately correlated with the chroma in each of CIE L*a*b*, CIE L*u*v* and YIQ color spaces, but is lowly correlated with the lightness and hue. Finally, the consistency among color attributes between CIE L*a*b* vs. CIE L*u*v* is highly significant. The consistency of lightness and chroma between CIE L*a*b* and YIQ, or CIE L*u*v* and YIQ, is moderately significant, but the hue is in the lower significance.