SmartColor: Automatic Web Color Scheme Generation Based on Deep Learning

Abstract

The color scheme plays an important role in different aspects of our everyday lives, such as web design and human-computer interaction. The generation of color scheme requires a long-term accumulation of design experience and advanced knowledge of color matching. However, there is little work focusing on the automatic generation of color scheme based on learning capabilities. In this work, we propose a novel color scheme designer, SmartColor, which incorporates deep learning methods with knowledge of color psychology. The Generative Adversarial Network (GAN) is used to learn experienced insights from widely recognized color schemes obtained from online color matching websites. Color schemes based on various themes are transformed as statistical constraints in the construction of the objective function of GAN. SmartColor is both data-driven and knowledge-driven. In contrast to current color scheme solutions. SmartColor will automatically create color schemes based on the input theme. Experimental results show that SmartColor was successful in creating color schemes for websites.