Classification of Text and Non-Text from Bilingual Document Images Using Deep Learning Approach

Abstract

In this work, we have presented an efficient approach for classification of text and non-text document information from real time office documents images printed/handwritten which are bilingual using a deep learning approach i.e., U-net architecture for experimentation purpose. We have created our own dataset containing 2000 document images. Initially pre-processing is applied on the input document images proposed method is compared with other existing methods and obtained accuracy of 99.62% different performance measure i.e., (Specificity, Sensitivity, Precision, F1-Score) used in the experimentation.