Real-time emotion recognition and interaction design of visual art images on mobile terminals based on deep learning

Abstract

Emotion recognition in visual artworks remains challenging because artistic images often contain abstract form, stylistic variation, and distorted visual representations that differ significantly from natural images. Existing deep learning models frequently struggle to generalize across artistic styles and accurately capture emotion-related cues; this study proposes a real-time visual art emotion recognition framework that combines a MHAI feature extraction module with transfer learning based on a pre-trained Inception-V3 network. The proposed architecture is designed to enhance the extraction of emotionally salient features while improving robustness to stylistic diversity in artworks. The framework incorporates image preprocessing, augmentation, attention-based feature learning, and emotion classification and was evaluated using the WikiArt emotion and ART500K datasets. Experimental results demonstrate classification accuracies of 98.5% and 96.7% on the WikiArt and ART500K datasets, respectively, with corresponding mAP values of 96.2% and 92.4%. The results indicate that the proposed attention-guided transfer learning strategy effectively captures emotional characteristics from heterogeneous visual art images and outperforms existing approaches. The developed framework offers potential applications in digital art analysis, intelligent learning systems, interactive museums, and emotion-aware human–computer interaction.