參數化色彩意象自動商品配色與分析

Automatic Product Color Scheme Using Parametric Color Images with Analysis

彭雅芳(Peng Ya-Fang)

指導教授 ： 周遵儒

國立臺灣師範大學/科技與工程學院/圖文傳播學系/碩士(2019年)

https://doi.org/10.6345/NTNU201901174

摘要

配色是商品設計過程中重要的一環，經常耗費許多時間溝通修改，然而時效性是今日幾乎所有商業活動的必備需求，因此即時的設計需求配色建議是亟待解決的議題。本研究應用小林重順在色彩意象尺度（Color Image Scale）（Kobayashi, 1991）中所列舉的色彩意象（Color Images）（Kobayashi, 2006）為設計參數，開發「色彩意象抽取演算法」，用來抽取描述語句中三個隱含的色彩意象，再收集這些意象所對應的大量色彩組合，透過「多意象色彩調和演算法」篩選出符合調和特性的部分，建立「參數化色彩意象配色系統」，使得所設計的商品在色彩呈現上具有描述語句所意圖傳達的色彩意象。
　　為驗證上述兩個演算法和整體系統的輸入輸出對應結果是否符合期望，本研究設計三組實驗，使用問卷調查法評估所設計方法的成效，三組實驗分別是：A、「描述語句輸入、色彩意象輸出」評估；B、「色彩意象輸入、商品配色輸出」評估；C、「描述語句輸入、商品配色輸出」評估，探討演算法與系統的精確度和召回率，結果如下：A實驗的精確度最高為70%（色彩意象輸出第一名即包含期望意象），而召回率最高為55%（前兩名色彩意象輸出包含期望意象），顯示僅使用「色彩意象抽取演算法」時建議可參考前兩名色彩意象輸出。B實驗及C實驗的的精確度皆為80%，「多意象色彩調和演算法」與「參數化色彩意象配色系統」能夠提供相對滿足語言和配色的轉換需求。設計背景與非設計背景的受測者在問卷結果的表現上沒有顯著的差異。總結來說，雖然在意象抽取的部分還有很大的進步空間，但在色彩意象與配色的對應上具有很高的可行性。

關鍵字

色彩意象 ； 色彩調和 ； 自動配色 ； 自然語言處理

並列摘要

Color scheme is an important part of the product design process. It often takes a lot of time to communicate and modify. However, timeliness is an essential requirement for almost all business activities today, so the proposal to provide color scheme immediately is an urgent issue. This study applies the color image (Kobayashi, 2006) listed in the Color Image Scale (Kobayashi, 1991) as a design parameter to develop a "color image extraction algorithm". The algorithm is used to extract three underlying color images from the description, and collect a large number of color combinations corresponding to these images, and pick out the parts that conform to the feature of color harmony through the "multi-images color harmony algorithm" to establish a "parameterized color image and color scheme system". Let the color of product have the color images that the description intends to convey.
　　In order to verify the effectiveness of the above two algorithms and the system, this study designed three sets of experiments, using questionnaires to evaluate the effectiveness of the methods, the three sets of experiments are: A, "description input, color image output “; B, "color images input, color scheme output"; C, "description input, color scheme output", to explore the accuracy and recall of the algorithms and the system, the results are as follows: Accuracy of experiment A is up to 70% (output the first color image contains the expected images), and the recall is up to 55% (output the top two color images contain the expected images). It is recommended to refer to the top two color images outputs when using the "Color Image Extraction Algorithm". The accuracy of both B and C experiments is 80%. The "multi-images color harmony algorithm" and the " parameterized color image and color scheme system " can provide the conversion requirements for language and color scheme. Subjects with design background and non-design background did not differ significantly in the performance of the questionnaire results. In summary, although there is still a lot to improve in the images of extraction, it is highly feasible in the correspondence between color images and color scheme.

並列關鍵字

Color Image ； Color Harmony ； Automatic Color Scheme ； Natural Language Processing

參考文獻

1. Beaird, J. (2010). The principles of beautiful web design. Melbourne, AU: SitePoint.

2. Chen, R., Hua, Q., Chang, Y., Wang, B., Zhang, L., & Kong, X. (2018). A Survey of Collaborative Filtering-Based Recommender Systems: From Traditional Methods to Hybrid Methods Based on Social Networks. IEEE Access 6(2018). 64301-64320. doi: 10.1109/ACCESS.2018.2877208.

3. Fraser, T., & Banks, A. (2004). Designer's color manual: The complete guide to color theory and application. San Francisco, CA: Chronicle Books.

4. Itten, J. (1986). The Color Star Stencils with Booklet. Canada, NA: John Wiley & Sons Inc.

5. Jahanian, A., Liu, J., Lin, Q., Tretter, D., O’Brien-Strain, E., Lee, S. C., Lyons, N., & Allebach, J. (2013). Recommendation system for automatic design of magazine covers. in Proc. IUI '13 Int. Conf, IUI Press (95-106).