AI-based color-matching system for cosmetic manufacturing

This study presents an AI-based automated color-matching system that uses a deep neural network (DNN) to replace manual iterative adjustments in color formulation. An on-site industrial dataset is collected from a cosmetic manufacturing system consisting of pigment and non-colorant material compositions, along with corresponding CIE L∗ a∗ b∗ (CIELAB) values. Based on this dataset, we develop a Sparsity-Aware Attention Network (SAAN) capable of predicting color outcomes under material variability and varying material combinations. By integrating the SAAN with an Automated Formulation Search Module, the proposed system is capable of automatically searching and assessing candidate formulations to match a given target color. The system is deployed in a real cosmetic manufacturing setting. Through this system, the cosmetics manufacturer reduced color-matching time and beaker-working iterations in an actual production environment. The system also reduce material and labour costs, thereby demonstrating clear practical benefits.