Exploring multi-tasking learning in document attribute classification

In this work, we adhere to explore a Multi-Tasking learning (MTL) based network perform document attribute classification such as the font type, size, emphasis and scanning resolution of image. To accomplish these tasks, operate on either segmented word level or uniformed size patches randomly cropped out document. Furthermore, hybrid convolution neural (CNN) architecture ”MTL+MI”, which is combination MTL Multi-Instance (MI) patch used joint for same attributes. The contribution paper are three fold: firstly, images patches, present full Secondly, propose MI (using words patches) combined CNN (“MTL+MI”) Thirdly, multi-tasking classifications and/or an intelligent voting system posterior probabilities each document’s attributes complete