Research on a Color Emotion Perception Method Integrating Unsupervised Clustering and Deep Learning

Abstract

This paper addresses the problem of color emotion perception by proposing an analytical method that integrates unsupervised clustering and deep learning. The aim is to automatically extract emotional information from color features and achieve interpretable modeling. First, statistical features of single-color images are extracted based on the HSV color space, and a regression model is constructed to predict pleasure scores. Comparative analysis is performed using various models, including linear regression, decision trees, random forests, k-nearest neighbor (KNN), and fully connected neural networks. The results show that nonlinear models and local modeling methods based on similar samples perform best. In the classification task, by constructing a binary label for "stimulation," the basic classification model can accurately characterize the psychological feeling of color in the arousal dimension. The deep learning model further demonstrates powerful feature extraction capabilities. Its convolutional feature map visualization results show significant differences in the local feature responses of different color samples. Second, K-means clustering is used to group color samples in an unsupervised manner, and semantic interpretation is performed based on emotion proportions. Color clusters are divided into types such as "calm and pleasant," "stimulating and active," and "energetic and pleasant," revealing distribution patterns of color in the dimensions of pleasure, calmness, and stimulation. Comprehensive analysis shows that brightness and saturation features significantly influence perceived pleasure, while stimulation and overall emotional atmosphere can be intuitively expressed through clustering results. The method presented in this paper not only achieves quantifiable modeling of color emotion perception but also provides data support for emotion-oriented color design and application, laying the foundation for future research on emotion prediction in multi-color combinations and complex scenarios.