使用透射率和反射率的水彩顏料數據於深度學習中預測半透明顏料混合物之模型

Abstract

學習混色對於新手來說是困難的。為了幫助新手畫家學習混色，我們提出一個可以預測半透明顏料混合物的模型，並且利用這個模型的預測結果去創建一個Smart Palette系統。首先，需要先建立一個用反射率與透射率表示兩種混色類別的水彩資料庫:分別是相同原色顏料混合的顏色份量增加與不同原色顏料混合的色彩混合兩類別。接下來，我們運用收集的資料，使用深度神經網路去訓練一個可預測半透明顏料混合物的模型。最後，為了讓使用者了解在真實生活中，如何使用兩個原色顏料如何混出目標色，我們創建了Smart Palette系統；當使用者從一張影像中拾取一個RGB色彩的像素，這系統會回傳它的混色配方$-$兩個原色顏料與它需要的使用份量。 半透明顏料混合物的模型預測的結果與真實混色資料的色差評估，測試集中有83%色差距離是小於5 (ΔE*ab < 5)。大於這個值域表示一個觀察者可以確定對照的色彩為兩個不同色彩。另外，在使用者研究評估報告中，首先，將使用者運用直覺、學習伊登色相環理論與使用Smart Palette配方進行混色的結果先採用色差距離公式(ΔE*ab)比較色差。接著，再使用t-test檢測這三個方法的色差距離的差異顯著性。最後，綜合色差距離與t-test的t值的結果,它可以證明使用Smart Palette配方進行混色的結果顯著地比其他兩者更接近目標色。因此，相較於傳統的混色學習方法,我們的系統(Smart Palette)的確可以有效幫助使用者學習更精準的混色。

Learning color mixing is difficult for novice painters. In order to support novice painters in learning color mixing, we propose a prediction model for semitransparent pigment mixtures and use its prediction results to create a Smart Palette system. Such a system is constructed by first building a watercolor dataset with two types of color mixing data, indicated by transmittance and reflectance: incrementation of the same primary pigment and a mixture of two different pigments. Next, we apply the collected data to a deep neural network to train a model for predicting the results of semitransparent pigment mixtures. Finally, we constructed a Smart Palette that provides easily-followable instructions on mixing a target color with two primary pigments in real life: when users pick a pixel, an RGB color, from an image, the system returns its mixing recipe which indicates the two primary pigments being used and their quantities. When evaluating the pigment mixtures produced by the aforementioned model against ground truth, 83% of the test set registered a color distance of ΔE*ab < 5; ΔE*ab, above 5 is where average observers start determining that the colors in comparison as two different colors. In addition, in order to examine the effectiveness of the Smart Palette system, we design a user evaluation which untrained users perform pigment mixing with three methods: by intuition, based on Itten's color wheel, and with the Smart Palette and the results are then compiled as three color distance, ΔE*ab values. After that, the color distance of the three methods are examined by a t-test to prove whether the color differences were significant. Combining the results of color distance and the t-values of the t-test, it can demonstrate that the mixing results produced by using the Smart Palette is obviously closer to a target color than that of the others. Base on these evaluations, our system, the Smart Palette demonstrates that it can effectively help users to learn and perform better at color mixing than that of the traditional method.