Sentence Generation System Using Affective Image

摘要

Various methods of sentence generation have been proposed for chatbot and short novel as examples. Some studies have attempted to generate sentences such as a caption automatically from picture image. Human beings can describe affective impressions on picture images. Therefore, it is important for robots to communicate with human beings about impressions of picture images. However, few studies have focused on an affective information of sentence generation. According to previous researchers, human beings evoke specific scenes or images when they listen to music and the scenes are described by lyrics of the music. Focusing on this phenomenon, we constructed a system that can generate sentences from picture images. Our system recognizes object of an input image and analyzes color information of an input image Then, our system proposes words related to the object and color impression to generate sentences we modeled Recurrent Neural Network. Our system can help user to do creative activities.