Kakao FE Deploys Internal Image Color Extraction Library Based on K-means

The Kakao FE Platform organization deployed an image color extraction library internally, developed to fill empty spaces in ad templates. Initially, the goal was to validate ad effectiveness through automatic image-based extraction before requiring advertisers to register colors separately.

K-means Clustering Application

It was difficult to extract natural representative colors from images with mixed colors using simple averages or modes. Therefore, the K-means Clustering algorithm was applied to group similar pixels based on RGB/HSV values and extract the color of the cluster containing the most pixels as the representative color. Optimal results were derived through centroid initialization and K-value configuration.

Storybook Utilization and Library Packaging

To facilitate efficient communication with designers, a Storybook-based testing tool was created to verify color extraction options in real time. Subsequently, A/B testing confirmed that background color filling had a significant impact on ad efficiency, and this logic was packaged and deployed as an internal library for reuse by other teams.