Natural scene statistics based publication classification algorithm using convolutional neural network

摘要

As digital devices which are capable of viewing contents easily such as mobile phones and tablet PCs have become widespread, publications are being digitized rapidly and the market for digital publications is growing up. However, digital publications are illegally distributed in the form of digital images. It is necessary to identify each digital image for protecting the copyrights of digital publications. We can detect the copyright infringement using publication identification algorithm after applying publication classification. In this paper, we suggest the publication classification method for mainly 4 types such as text, cartoon, webtoon, and regular picture. We use 2-layered CNN for publication classification using histogram images, which are extracted by NSS(Natural Scene Statistics), which usually is used for figuring out distortion in a natural image. We expect that our proposed method will be useful for protecting copyrights of publications.