Movie Genre Classification based on Poster Images with Deep Neural Networks
Wei-Ta Chu, Hung-Jui Guo · 2017
We propose to achieve movie genre classification based only on movie poster images. A deep neural network is constructed to jointly describe visual appearance and object information, and classify a given movie poster image into genres. Because a movie may belong to multiple genres, this is a multi-label image classification problem. To facilitate related studies, we collect a large-scale movie poster dataset, associated with various metadata. Based on this dataset, we fine-tune a pretrained convolutional neural network to extract visual representation, and adopt a state-of-the-art framework to detect objects in posters. Two types of information is then integrated by the proposed neural network. In the evaluation, we show that the proposed method yields encouraging performance, which is much better than previous works.