.png){kind=icon}
Return
AttDNet: An attribute-driven deep learning network for visual emotion analysis
.png){kind=icon}
PRE
AI
DOI：10.1016/j.patcog.2025.112098.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
With the rapid development of multimedia technology, more and more people tend to express their views through images on the Internet. Therefore, visual emotion analysis (VEA) has attracted significant attention from scholars all over the world. However, emotion is a highly subjective concept that can be influenced by various attributes of the image, such as color, brightness, faces, objects, scenes and so on. These attributes are associated with the process of human cognitive emotion and they are present not only in the global features, but also in the local features. However, most existing methods rely solely on global features for classification. Some methods focus on shallow local features, but the obtained information is not comprehensive enough. In order to comprehensively extract the emotion-related attribute information in images, we propose an Attribute-Driven Deep Learning Network (AttDNet). This network comprises an attribute branch with multiple modules and a facial branch that can obtain facial attribute. The modules in the attribute branch focus on various attribute information in the image. Among them, CSCAM module can extract color information and key objects in the image, LEconv module can extract context information of objects in images, and MLPooling module can extract scene information in images. In addition, to bring samples of the same emotional category closer together, we use CenterLoss to constraint the model. The experimental results demonstrate that the proposed method achieves state-of-the-art accuracy on public datasets.
Keywords:
visual emotion analysis
attribute-driven deep learning
color information
facial attributes
scene context
Journal
IF:
7.6
Papers:
1.3W
Citations:
4.5W
Organization
Cited Papers
No cited papers available