.png){kind=icon}
.png){kind=icon}
Return
Generating affective maps for images
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-017-5289-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Affective image analysis, which estimates humans' emotion reflection on images, has attracted increasing attention. Most of the existing methods focus on developing efficient visual features according to theoretical and empirical concepts, and extract these features from an image as a whole. However, analyzing emotion from an entire image, can only extract the dominant emotion conveyed by the whole image, which ignores the affective differences existing among different regions within the image. This may reduce the performance of emotion recognition, and limit the range of possible applications. In this paper, we are the first to propose the concept of affective map, by which image emotion can be represented at region-level. In an affective map, the value of each pixel represents the probability of the pixel belonging to a certain emotion category. Two popular application exemplars, i.e. affective image classification and visual saliency computing, are explored to prove the effectiveness of the proposed affective map. Analyzing detailed image emotion at a region-level, the accuracy of affective image classification has been improved 5.1% on average. The Area Under the Curve (AUC) of visual saliency detection has been improved 15% on average.
Keywords:
Affective map
Affective image classification
Visual saliency
Multiple instance learning
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W