Recognizing the Order of Four-scene Comics by Three-Path Convolutional Neural Networks

Recently, the comic analysis has become an attractive research topic in the artificial intelligence fields as comic engineering. In this study, we focused on the four-scene comics and applied deep convolutional neural networks (DCNNs) to those data for understanding the order structure. We proposed the novel approach for that problem by three-path DCNN with special input data formats. The hyperparameters of three-path DCNN are obtained by evolutionary deep learning (evoDL). The effectiveness of the proposed method is confirmed by computer simulations taking a real four-scene comics structure recognition problem as an example.