关键词: 图像情感, 迁移学习, 卷积神经网络

Abstract: Images are powerful tools with which to convey human emotions, with different images stimulating diverse emotions. In this paper, a data augmentation method based on small data sets is adopted. The advanced features （object category and deep emotion feature） are used, which are automatically extracted by the deep network combined with low-level features （color and texture features） to recognize emotion of the image. A high-level semantic descriptive phrase including compound emotions and object is output. The results show that the proposed method is superior to other traditional manual extraction methods or existing deep learning models and achieves 66.54% accuracy on emotion recognition on IAPS and GAPED data sets.

Key words: image semantics, transfer learning, Convolutional Neural Network（CNN）