This study proposes a fusion feature extraction and transfer learning-based framework for emotion recognition in art and design. We extract low-level visual features (color and texture) and semantic features, and integrate them using a multi-branch convolutional network. Transfer learning techniques are employed to enhance recognition performance on small-scale artistic datasets. Experimental evaluations on the Flickr dataset demonstrate that our method improves recognition efficiency by 8% and increases accuracy by 5% compared to traditional approaches. The proposed approach shows strong generalization across diverse artistic styles, offering a promising solution for emotion analysis in creative industries.
B. J. Liu (Thu,) studied this question.