Article: Emotion recognition in artistic images based on feature fusion and transfer learning Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.44 pp.1 - 17 Abstract: Currently, artistic images are scarce with limited sample sizes, and most sentiment analysis relies on low-level image features with low accuracy. To address this, this paper first extracts two-dimensional features from images in different colour spaces. It then employs multi-scale convolutional kernels to extract deep semantic information from images, fusing feature information from different dimensions to effectively preserve semantic features across scales. Finally, the transfer component analysis algorithm is employed to reduce dimensionality of features in source and target domains within original space. An improved joint subspace learning method is used to learn a feature transformation subspace, reducing the conditional probability distribution distance between source and target domains while balancing recognition accuracy across categories. Model optimisation is achieved through adversarial training. Experimental results demonstrate that the proposed model improves recognition accuracy by at least 3.82%, effectively enhancing the accuracy of emotional recognition in artistic images. Inderscience Publishers - linking academia, business and industry through research

Title: Emotion recognition in artistic images based on feature fusion and transfer learning

Authors: Laohui Liang

Addresses: Faculty of Marxism, Guangdong Mechanical and Electrical Polytechnic, Guangzhou, 510000, China

Abstract: Currently, artistic images are scarce with limited sample sizes, and most sentiment analysis relies on low-level image features with low accuracy. To address this, this paper first extracts two-dimensional features from images in different colour spaces. It then employs multi-scale convolutional kernels to extract deep semantic information from images, fusing feature information from different dimensions to effectively preserve semantic features across scales. Finally, the transfer component analysis algorithm is employed to reduce dimensionality of features in source and target domains within original space. An improved joint subspace learning method is used to learn a feature transformation subspace, reducing the conditional probability distribution distance between source and target domains while balancing recognition accuracy across categories. Model optimisation is achieved through adversarial training. Experimental results demonstrate that the proposed model improves recognition accuracy by at least 3.82%, effectively enhancing the accuracy of emotional recognition in artistic images.

Keywords: artistic image emotion recognition; feature fusion; transfer learning; adversarial training; feature extraction.

DOI: 10.1504/IJICT.2025.150409

International Journal of Information and Communication Technology, 2025 Vol.26 No.44, pp.1 - 17

Received: 07 Sep 2025
Accepted: 20 Oct 2025
Published online: 12 Dec 2025 *

Title: Emotion recognition in artistic images based on feature fusion and transfer learning

Keep up-to-date