Emotion based image musicalization

Playing appropriate music when watching images can make the images vivid and bring people into their intrinsic world. In this paper, we propose to musicalize images based on their emotions. Most of previous works on image emotion analysis mainly used elements-of-art based low-level visual features, which are vulnerable to the arrangements of elements. Here we propose to extract visual features, inspired by the concept of principles-of-art, to recognize image emotions. To enrich the descriptive power, a dimensional perspective is introduced to emotion modeling. Experiments on the IAPS dataset demonstrate the superiority of the proposed method in comparison to the state-of-the-art methods for emotion regression. The music in MST dataset with approximate emotions to the recognized image emotions is selected to musicalize these images. The user study results show its effectiveness and popularity of the image musicalization method.