Establishing colour harmony evaluation and recommendation model for clothing colour matching based on machine learning and deep learning

Abstract

Appropriate colour combinations improved aesthetic design quality and provided a comfortable and pleasant visual experience. However, applying current colour harmony models to clothing colour matching raised doubts about whether the existing theory needed to be refined, requiring further clarification with modern aesthetic perspectives. The study conducted a psychophysical experiment to investigate modern people's perceptions of colour harmony in clothing colour combinations. The results indicated that modern perceptions of colour harmony in clothing differed significantly from previous theories. For applications, two colour harmony models were established with unified fashion datasets for evaluating clothes matching based on colour harmony rules and observers' perceptions. The result showed that the rule-based model could accurately predict all items following nine colour harmony theories, and the perception-based colour harmony evaluation models aligned with contemporary aesthetic preferences were developed with a semi-supervised learning approach. The models, including support vector machines and custom convolutional neural networks, could predict colour harmony perception for the input images with high performance, and the model based on the generative adversarial network could provide colour recommendations for colour matching. The machine learning and deep learning model proposed in the study could be used for aesthetic judgment to generate clothing colour recommendations and provide valid design suggestions for clothing colour matching in the fashion and clothing industry.