Abstract
隨著電腦多媒體時代的來臨,影像的處理、顯示及列印已由黑白昇級至 彩色。於是,電腦處理彩色訊息的依據-人類彩色訊息處理模式便迫切需 要建立。本質上,彩色較之黑白不但多了兩維度的客觀訊息(適於樣型識 別等應用),並且具有黑白所欠缺的主觀訊息而更適於美感判斷等應用。 因此,相對於傳統彩色影像處理僅考慮RGB訊息,本論文提出一個“色彩 多元關係”(見論文英文本第2頁之Fig.1.1)來更完整地描述人類色彩知覺 的主客觀多元特性。此關係由外而內涵蓋了:頻譜(Spectrum)、色彩外 觀(Color Appearance)、色彩意象(Color Imagery)、色彩美感(Color Harmony)、色彩選擇(Color Selection)及色盲(Color Blindness)。應用 此“色彩多元關係”將可建立一套“電腦色彩諮詢系統”(見論文英文本 第3頁之Fig.1.2)來答覆色覺正常者與色盲者日常生活遭遇的色彩相關問 題。色彩外觀強調人眼所見之色彩皆經由光源、背景及周遭環境調節而得 ,其與儀器測量小範圍區域所得未經調節的資料不同。其典型現象-共存 性色對比是一種前景顏色受其背景之影響而偏向背景顏色之補色的色彩調 適作用,此色覺偏移之方向及大小則與前景、背景之顏色及幾何面積有密 切關係。傳統上,此色覺偏移係藉由經驗設定Hunt色彩模型之 p參數(-1< p<0)求得。為改善經驗設定 p參數的不確定性,本論文提出兩種幾何影響 因素:前景色與背景色之“面積比”及其“分隔距離”來量化 p參數(見 論文英文本第29頁之Equation 3.2),其預估結果並與實地量測之數據相 比較。結果顯示:對色覺偏移之方向可作精確的預估,而對色覺偏移大小 的預估則稍遜,平均誤差(於CIE1976UCS色彩空間中)為0.3485%。 此 外,色彩外觀強調環境調節的特性亦極適合應用於跨媒體的色彩預覽。相 較於傳統利用媒體色域校正(Gamut Mapping)方式達成不同媒體間儀器色 彩量測值的一致,本研究提出一跨媒體複雜影像評估系統(見論文英文本 第40頁之Fig.3.12),擬評估四種色彩(外觀)模型(von Kries、CIELAB、 RLAB及Hunt)於彩色螢幕上預覽彩色印表機列印稿於燈箱照明下之色彩。 與相關研究之差異在於螢幕及燈箱之光源亮度不同及周圍環境微亮(較接 近實際應用情況)。結果顯示,Hunt色彩模型較其他模型更符合不同媒體 間預覽複雜影像的需求。色彩美感意謂人類看見二種以上色彩時所產生的 滿意感受。傳統色彩美感之判斷以主觀經驗為依據,本論文提出的色彩美 感評估方法為(見論文英文本第56頁之Fig.4.2):先根據色彩在特定色彩 空間中的形態距離(Morphological Distance)計算其關於“熱鬧色”與“ 寧靜色”的隸屬度(對室內色彩設計之應用而言),接著根據色彩-語意之 關連特性設計色彩語意軸“熱鬧(Exciting)-寧靜(Calm)”,以將人類對 彩色影像之心理印象於此軸上表示成“色彩意象分佈”(色彩意象意謂人 類看見色彩時所產生能以言語表達的感受,即心中印象)。而根據輸入影 像之色彩意象分佈與代表社會整體美感趨勢之“美感參考分佈”間的類似 性,色彩美感之程度(分數)便可獲得。實驗顯示,電腦計算的色彩美感分 數與69位受測者之問卷結果有高達95.56%的關聯性(Correlation) ,遠優 於傳統Moon-Spencer理論之結果(50.73%)。此外,所用的色彩語意軸及色 彩意象分佈不但能描述流行趨勢,而且能解決72.3%的色彩美感問題,此 亦優於傳統Moon-Spencer理論的61.1%,及面積反比理論的34.6%。傳統的 色彩選擇方式除了耗時的人工手動外,選色依據以客觀的色差大小最普遍 ,但所選出的色彩僅能適合樣型識別等客觀應用,無法滿足美感應用的主 觀需求。本論文所提出之美感色彩的選擇方法為(見論文英文本第95頁之 Fig.5.3):將輸入之彩色影像先編碼為色彩意象(方法同以上所介紹之色 彩美感評估),再根據代表社會整體美感趨勢之“美感參考分佈”進行色 彩意象修改,最後將修改後之色彩意象解碼回復為RGB色彩值以顯示於螢 幕上。為適用此意象編碼、意象修改、意象解碼之色彩選擇模式,我們提 出四種“美感參考分佈”:極端寧靜(Extremely Calm)、相當寧靜(Quite Calm)、稍微寧靜(Slightly Calm)與實際(Practical)分佈。實驗顯示, 採用“相當寧靜”意象分佈選色後之影像,其美感分數的改善可達四倍。 色盲為先天基因缺陷造成生理上色彩感測或色差處理機制發生病變而導致 色覺異於常人的現象,故色盲者好比生活在一個“變了色的世界”之中。 色盲依其生理病變程度可分為二大類:色弱(三種色彩感測細胞皆異常, 佔總人口6.3%)與單色色盲(一種色彩感測細胞或色差處理機制缺乏、剩餘 之二種色彩感測細胞異常,佔總人口2.133%)。色盲者之色覺異常現象極 易造成對色盲表之觀測、交通號誌之辨識、及(手術時)動靜脈血管之分辨 有困難。臨床上至今對色盲僅能進行檢測,尚無有效的治療方法。本論文 提出二種分別基於色彩三刺激(Tristimulus)特性與色彩頻譜特性之方法 :三刺激法針對色盲色覺之模擬與等化利用調整Hunt色彩模型中相對應之 生理三刺激或色差參數來將色彩模型轉變為色盲模型以進行色盲色覺模擬 ,並結合其反函數為色彩修正機制進行色盲色覺等化以修正色盲色覺(見 論文英文本第121頁之Fig.6.4);頻譜法則進一步推導“光學式色盲色覺 模擬鏡片”之頻譜穿透率(見論文英文本第138頁之Equation 7.13),以作 為未來製作光學鏡片之基礎。實驗顯示,當三刺激法所模擬之色盲表影像 與原始影像並列時,單色色盲幾乎分辨不出;經等化修正後,單色色盲則 可分辨出正確的色盲表數字。頻譜法之單色色盲模擬結果雖稍遜於三刺激 法(因所推導之頻譜穿透率包含精神物理學實驗的量測誤差) ,但頻譜法 可適用於佔色盲大多數之色弱,甚至可應用於模擬部份動物、昆蟲(如: 金魚、鴿子、猴子及蜜蜂)之“色覺”。藉由以上的研究,我們已對“色 彩多元關係”加以分析並得到了具體的成果-“電腦色彩諮詢系統”,可 針對色覺正常者與色盲者進行:色彩外觀預測、色彩美感判斷、美感色彩 建議、色盲色覺模擬與等化諸功能。為適合不同媒體間色彩顯示的需要, 處理產生之色彩可藉跨媒體色彩預覽功能於彩色螢幕上預覽彩色印表機將 輸出之色彩。 A successful modeling of human color processing system plays animportant role for computer-based color evaluations. Conventionally, acomprehensive (both objective and subjective) color modeling from anyengineering perspective has received limited attention because of itsinvolving the complex characteristics of the physics (e.g., spectralstimuli), the physiology (e.g., color blindness), the psychophysics(e.g., color appearance), and the psychology (e.g., color imagery, colorharmony and color selection). In this study, a comprehensive colorrelationship is proposed to clarify those transforms from the input ofspectral stimuli into color appearance, color imagery, color harmony,color selection, and even color blindness.Color appearance is the color vision indicating the physiologicaladaptation to both background and viewing conditions. For a typicalphenomenon of simultaneous color contrast, the empirical "p factor"(ranged from 0 to --1) conventionally used in color appearance models isquantified by both the number of pixels and the separation betweenforeground and background colors. This quantizationreports a 0.3485% (CIE1976UCS) color-difference between the predictedand measured results. Besides, the color appearances between softcopyand hardcopy are predicted by four color models (i.e., von Kries, CIELAB, RLAB and Hunt) under the more general environments of unequalluminance levels and dim surround. Results of the Zscale values measured by the psychophysical procedure based on 30ordinary observers demonstrate that, on the average, the Hunt modelachieved the best color prediction against the worst one of the RLABmodel. Color imagery and color harmony are closely related to the psychologicalfeature of human pleasure. A new pleasure-related function of CLD(colorlinguistic distribution, corresponding to the color imagery) is proposedto quantitatively represent human mental color impression of interiorimages upon a designed one dimensional linguistic-based image scale of" EXCITING-CALM". Supported by a database, thedistribution state of CLD is capable of indicating the fashion trends inTaiwan. Also, on the image scale, the grade of color harmony can bemeasured by its CLD similarity to an RHD(reference harmonydistribution). Evaluated results demonstrate a 95.56% correlationcoefficient with those questioned from 69 young people, which is betterthan that (i.e., 50.73%) of the famous Moon-Spencer theory based on huefeature. Color selection's performance depends chiefly on the focused feature andapplied fields. A novel technique of color selection considering colorharmony is proposed for subjective-orientedapplications. Based on encoding color image into itspleasure-related feature of CLD, the selection of colors in thetristimulus color space can correspond to the specification of CLD uponthe linguistic- based image scale. Also, the specified CLD can be decodedto tristimulus values by a reverse procedure. This design can gain 400%harmony grades over the original image. Color deficiencies inflict different color visions on almost 8.438% ofthe entire population (i.e., 6.3% for the type of "anomaloustrichromatism", 2.133% for the type of "dichromatism", and 0.005% forothers). Two potential approaches based on tristimulus values andspectrum are constructed. The tristimulus approachspecifies the related tristimulus parameters of color appearance modelto zero for representing the absence of physiological color-differencesignal to report dichromatic color-blind simulation. Also, thisspecification can be applied to the lack of physiological cone sensors(e.g., [Nature, 1995;95vienot Science, 199595david and 199696neitz]).Moreover, the spectrum approach derives the spectraltransmittance of an optic-based simulator from the reciprocal ofwavelength discrimination functions measured experimentally. In theconsideration of dichromatic population ratio, this derivation reports asimilar performance to the tristimulus approach, and is suitable to boththe dichromatic and the anomalous trichromatic color deficiencies. Finally, a concrete application of the comprehensive color relationshipis developed, called "computer-based color consultation system", forevaluating colors objectively and subjectively from the viewpoints ofcolor-normal and color-defective people.