Image Aesthetic Assessment Based on Emotion-Assisted Multi-Task Learning Network

Image emotion recognition and image aesthetic assessment are recent research hotspots in user perception of image content. However, for the study of image aesthetics and image emotion, the vast majority of studies are separated from the relationship between the two, respectively. But in fact, there is a potential and explicit connection between image aesthetics and image emotion. In this paper, we use a multi-task convolutional neural network. To achieve the purpose of this experiment, for the first time, we propose a strictly labeled dataset, which contains multi-person annotated aesthetic scores and eight types of emotion distributions. With the help of this dataset, on the basis of aesthetic assessment based on scene and object branch, the method of an emotion-assisted multi-task learning network is adopted to improve the performance of aesthetic assessment. The experimental results show that our network structure achieves excellent performance in many indexes of image aesthetic assessment when comparing several state-of-the-art aesthetic assessment algorithms and convolution neural networks with strong performance.