Colors have a significant impact on marketing so designers often use color strategically to shape brand personality and purchase intent. The automatic color design of advertising images is challenging because it must satisfy esthetic goals while also complying with the color design for different content, brand or targeted audience. This paper proposes a data-driven method to color advertising images automatically according to input product color and context-related keywords. To generate good coloring suggestions, we collected 13,000 well-labeled adverting images and built two probabilistic models to capture stylistic color properties. To collect the color preferences of potential users, we developed a color analyzer to display the result of clustering advertising images. Using the visual display, a user can explore and select images to guide our model to generate more personalized colors. We applied our method to three coloring tasks including keyword-based coloring, product-based coloring and user-guide coloring to test its performance against three coloring goals. In perceptual studies, colorings generated by our method were preferred to those generated by other models or created by non-professional students.

颜色对营销有重大影响，因此设计师经常策略性地使用颜色来塑造品牌个性和购买意图。广告图像的自动色彩设计具有挑战性，因为它必须满足审美目标，同时还要符合针对不同内容，品牌或目标受众的色彩设计。本文提出了一种数据驱动的方法，根据输入的产品颜色和上下文相关的关键字对广告图像进行自动着色。为了产生良好的着色建议，我们收集了13,000个标签良好的广告图像，并建立了两个概率模型来捕获样式颜色属性。为了收集潜在用户的颜色偏爱，我们开发了颜色分析器来显示广告图像聚类的结果。使用视觉显示，用户可以浏览并选择图像以指导我们的模型生成更多个性化的颜色。我们将我们的方法应用于三种着色任务，包括基于关键字的着色，基于产品的着色和用户指南的着色，以针对三种着色目标测试其性能。在感知研究中，通过我们的方法生成的颜色优于其他模型生成的颜色或非专业学生生成的颜色。