CNN-Based Emotion Classification in Visual Art for Therapeutic and Creative Applications

Emotion recognition from artworks has the potential to enhance the experience of art exhibitions, where emotions conveyed by artworks can enhance the viewer’s experience with synchronised lighting, music, and multimedia elements. Integrating emotion detection technology and applications to the art experience enlarges the way of perceiving and embracing art, leading to personalized therapy applications (e.g., art therapy). We used Convolutional Neural Networks and Transfer Learning to detect emotions in paintings, comparing three state-of-the-art models with different characteristics. A prototype application has been developed to show the classification capability of the best-performing model. The results highlight the effectiveness of our approach, particularly for binary classification, in real-world applications, such as adaptive art exhibitions and real-time art therapy tools. Challenges, such as dataset limitations and the subjective nature of emotions in art, were addressed through careful dataset integration and preprocessing, as well as the use of transfer learning to optimize performance. This work introduces applications of CNN in art therapy, immersive art experiences, and beyond, by demonstrating the potential of combining datasets and applying advanced deep learning techniques to emotion recognition in art, from enhancing art experiences to supporting emotional analysis in other creative industries. Keywords-Emotion detection; CNN; Transfer lea