Semantic and visual classification of digitized documents

This paper presents an overview of the problem of automatic classification of digitized documents. We will see the options available to describe both the visual appearance and the textual and semantic con-tents of these documents. We will review how these descriptions can be used for the classification, clustering or retrieval of digitized documents. We will summarise the state-of-the-art approaches both from the computer vision and natural language processing fields and will see how the latest breakthroughs in Deep Learning have revolutionized these fields.