模棱两可
人工智能
计算机科学
深度学习
卷积神经网络
情感表达
表达式(计算机科学)
创造力
前提
可视化
特征(语言学)
特征提取
情感计算
数字艺术
人工神经网络
心理学
特征学习
计算创造力
感知
人机交互
模式识别(心理学)
情绪识别
视觉感受
数据可视化
认知心理学
作者
Prashant B. Wakhare,Riyazahemed A. Jamadar,Sanjay Bhilegaonkar,Pallavi V. Mulmule
出处
期刊:ShodhKosh Journal of Visual and Performing Arts
[Granthaalayah Publications and Printers]
日期:2025-12-20
卷期号:6 (3s)
标识
DOI:10.29121/shodhkosh.v6.i3s.2025.6952
摘要
The expression of emotion is a characteristic but difficult feature of digital art that is most frequently expressed in abstract visual features instead of direct semantics. This paper explores how deep learning methods can be used to learn and analyze emotional expression in digital artwork. The proposed hybrid model that integrates Convolutional Neural Networks and Vision Transformers will be able to capture local visual features, including color and texture as well as global compositional structure. A selected collection of various digital artworks is modeled and cited by a hybrid emotion system based on discrete categories and dimensional valence-arousal models. The experimental findings prove that the proposed hybrid method is more successful than CNN and transformer baselines on both emotion classification and regression problems with a higher F1-score, reduced error in prediction, and increased correlation with human emotional ratings. Embedding-level and qualitative analyses also indicate that the learned representations are able to maintain emotional continuity as well as ambiguity in artistic expression. The results affirm that emotion in digital art is multidimensional and optimal with regard to integrated local-global feature learning. The presented work contributes to the development of affective computing in the world of creativity and offers a premise to the study of emotional art, curating it, and creative collaboration between humans and AI.
科研通智能强力驱动
Strongly Powered by AbleSci AI