Examination Paper Image Segmentation with Adversarial Network

Examination paper analysis is important for students to improve their learning efficiency. Traditional paper examination papers are difficult to sort out, which makes collecting mistakes for future review both time-consuming and laborious, either by handwriting or existed software tools. To easy the process, we propose a layout analysis method combined with the conditional generative adversarial network (CGAN). The traditional semantic segmentation structure is improved and used as a generator in the network, while a discriminator is designed to make the segmentation results more accurate. The motivation is that the discriminator can judge the authenticity of the image, so it can help reduce the unreasonable phenomenon in the semantic segmentation results of the generator. The experimental results show that by this method the examination paper image can be welly split into various components, which provides convenience for further sorting and analysis of the examination paper images.