Flexible neural color compatibility model for efficient color extraction from image

Abstract

Color choice is an essential aspect of many applications, including graphic design, web design and fashion design. The selection of colors can have a significant impact on the overall aesthetic and appeal of a design, as well as its effectiveness in conveying a particular message or mood. This paper introduces new and simple tools for choosing colors. First, we introduce a convolutional neural network that scores the quality of a set of five colors, called a color theme. Such a network can be used to rate the quality of a new color theme. Second, we propose a method to extract a variable-size palette from an image. The size of the extracted palette can vary depending on the color richness of the image. Third, we demonstrate simple prototypes that apply the trained neural network and the palette extraction method to tasks in graphic design, such as improving existing themes. Our proposed network has the advantage of being significantly simpler than other state-of-the-art methods with better performance.