Establishing a Large Scale Dataset for Image Emotion Analysis Using Chinese Emotion Ontology.

With the development of visual social network, more and more people like to present themselves using images or videos. Visual emotion analysis is becoming one of hot research topics. According to Jou’s idea [1], emotion presentations of the Western and the Eastern are much different due to the culture difference. There are some popular emotion models such as Plutchik’s model and so on. And there is not one-to-one correspondence between these emotion models. All of the existing image databases for emotion analysis are built by Plutchik’s and Mikels’s emotion models. However, most researches on Chinese text emotion analysis used Xu’s model [2]. And there are not the corresponding image datasets for emotion analysis with Xu’s model. In this paper we establish an image dataset for emotion analysis by collecting images from Flickr using Chinese Emotion Ontology of Xu’s model. In addition, we design a dataset refinement (de-noising) strategy to promote the confidence of emotion labels for the images. Finally, we establish the dataset CH-EmoD which includes a sub dataset with single emotion label and a sub dataset with multiple emotion labels. Furthermore, we provide the baselines of emotion classification and multi-label emotion classification by using state-of-the-art emotion/sentiment classifications algorithms Alexnet [3] and PCNN [4]. The experimental results demonstrate that the dataset works well on emotion classification and multi-label emotion classification and the proposed dataset refinement strategy is effective.