花語之色彩意象應用於色彩建議與分析

Abstract

本研究所側重分析的焦點以日本學者小林重順建立色彩意象座標（Color Image Scale）與色彩意象詞彙資料庫（Color Image Word Database），讓情緒與色彩或色彩組合標準化、數值化，以奠定學理討論基礎，其利用語意差異法度量色彩及意象的關聯，與日本色彩與設計研究所（Nippon Color and Design Research Institute, NCD）合作開發色彩意象座標。透過自然語言處理（Natural Language Processing, NLP）技術將一般口語化的表達轉換至專業的一個或多個設計參數的辨識，用於人工智慧（Artificial Intelligence, AI）深度學習（Deep Learning）訓練出符合大數據內容呈現趨勢優化的色彩建議的方法，提出具體建議。透過設計3組實驗「多意象色彩調和演算法」、「色彩意象抽取演算法」、「花卉圖片重點色彩擷取」，進行提取3色色彩組合當作已知色，實作於「色彩建議演算法」輸出建議色，利用網路問卷調查分析滿意度，結果顯示色彩建議後的5色色彩組合的滿意度平均數都比4色色彩組合高。本研究的主題花語之色彩意象應用於色彩建議後的4色、5色色彩組合的滿意度平均數均達3分以上具有正面的評價。另外，本研究觀察審美度方程式M=O/C，花卉圖片重點色彩應用於色彩建議後的4色、5色色彩組合，都有100%符合M>0.5，發現應是花的顏色色相大多較為相近，產生對應到的數值不會差太大的現象，在曼賽爾色彩系統中如果O與C的落差不夠大，計算得到的數據就不會差太大，進而發現當色彩色相都較為相近時只採用審美度來進行評量色彩調和度是不夠的。 花語被加以利用於色彩意象的表現，輔助設計半自動化色彩建議方法，產生具有代表性或獨特性的色票，未來得以應用於印刷與設計產業中，解決一般非專業人員色彩運用能力不足的困境。COVID-19疫情觸動數位轉型契機，迫切需要大量的資訊傳遞、搜尋與雲端儲存及大數據的使用。科技的進步讓科技推動模式逐漸由技術轉為需求導向（陳聖智，2021），色彩建議方法的效能與創新應用的可行性，導入人工智慧概念，無須透過漫長歲月經驗累積養成，輔助更多有設計需求但能力不足的人，即時性設計因應少量多樣、個人化、個性化的趨勢設計潮流，亦是本研究主要課題以供後續相關研究與應用之參考。

This study is focused on the Color Image Scale and the Color Image Word Database, which was established by the Japanese scholar Shigenobu Kobayashi as a standardized and quantitated platform of emotions and colors or combinations thereof for theoretical discussions. Mr. Kobayashi collaborated with the Nippon Color and Design Research Institute, utilizing the linguistic differentiation method to measure the relationship between a color and its color image and develop the Color Image Scale. This study uses natural language processing technology to transform colloquial expressions of flower symbolism into one or more identifiable professional design parameters and then trains the big data by artificial intelligence-based deep learning in order to provide practical suggestion and optimized tendency for color usage.In this study, through 3 groups of experiments:"multi-image color harmonization algorithm", "color image extraction algorithm" and "accent color capture of flower pictures", the 3-color combinations were extracted as known colors, and the aforesaid color suggestion algorithms were used to output suggested colors. After analyzing the satisfaction survey conducted by online questionnaires, the results showed that the arithmetic mean of satisfaction survey of the 5-color combinations, which use the suggested colors, was higher than that of the4-color combinations. The arithmetic mean of satisfaction survey of the 4-color and 5-color combinations, all of which use suggested colors generated by color image of the flower symbolism at issue, was more than 3 points and thus possessed a positive evaluation. This study further reviewed the aesthetic measure equation: M = O / C and found that 100% of the 4-color and 5-color combinations, all of which use suggested colors generated by the accent colors of flower pictures, were resulted in M> 0.5. It was likely because the color hues of flowers were substantially similar, and therefore the corresponding ratios did not differ too much. In the Munsell color system, if the differences between O and C are not large, the calculated ratios will not differ too much. Hence, it is concluded that merely using the aesthetic measure equation to evaluate color harmony is insufficient when the color hues are relatively similar. This study demonstrates that flower symbolism, if used in the expression of color image to develop a self-generating color suggestion method, can create representative or unique color tickets. It can be applied to the design industry in the future, solving the dilemma faced by laypersons who will not need to accumulate long-term experience. By introducing artificial intelligence into the printing and design industries, the color suggestion method, with its efficiency and applicability, can be a technological innovation and a new model of design service and a reference for subsequent researches and applications.