Comics image processing
Nina S. T. Hirata, Igor dos Santos Montagner, Roberto Hirata · 2016
We employ an image operator learning method to segment text in comic images. Since the method is based on learning from pairs of input and corresponding expected output images, it is flexible with respect to alphabet sets and text orientation. The method is applied on both Japanese and European comics. Results indicate that most text regions can be straightforwardly identified from the output images.