Finding Theme Color Palettes

Abstract

This project is about finding theme color palettes from images of famous paintings. A color palette, in the digital world, refers to a full range of colors that can be displayed on a device screen or other interfaces, or in some cases, a collection of colors and tools for use in paint and illustration programs. We can use the methods we present in this project to get color palettes from images we love which can be an interesting process. In our daily life, the theme color palettes can be used for clothing matching, interior design, and even plate presentations. These applications can let people’s aesthetic taste and quality of life improve. Additionally, this project will enable more non-professionals to have more professional color perceptions. Three different kinds of color quantization algorithms (k-means, median-cut, octree) are implemented to find theme color palettes of reference images in RGB and L*a*b* color spaces. We then use four objective image quality assessments (PSNR, SSIM, VIF, GMSD) to evaluate the fidelity of quantized images with the original images. According to our experimental results, the color palettes obtained by different color quantization algorithms are different, and the k-means algorithm has the best performance in both RGB and L*a*b* color spaces, and the color space conversion (RGB to L*a*b* color space) improves the visual fidelity of images.