Emotion Recognition in Art Creation Using Visual Image Analysis Techniques

Emotion drives artistic creativity and expression. Artists begin with emotional intent and convey it through visual language, which then evokes emotional responses in viewers. To support this transformation, educators need effective analytical methods. This study examines emotional recognition in artworks using visual image analysis. A dataset of 500 art images representing six emotions—affection, friendship, love, homesickness, patriotism, and sadness—was used. Images were collected from public sources and independently rated by three art experts, achieving strong agreement (Cohen’s κ = 0.87). The dataset was split into 70% training (350 images) and 30% testing (150 images), with balanced emotion categories. All images were resized to 256×256, converted to grayscale, and normalized before feature extraction. Among the tested methods, PCA performed best, achieving 94.5% exactness, 95.9% recall, 95.9% accuracy, and 97.6% precision. It was followed by LDA, stepwise regression, and a deep learning model. PCA showed the highest average accuracy (0.8567), with LDA close behind, while stepwise regression and the deep learning model reached 0.803 and 0.823. Both PCA and LDA produced low error rates (under 0.1). Overall, PCA and LDA effectively identify emotional patterns in artworks and support deeper understanding of how visual structure and composition convey emotion.