Text Line Segmentation for Challenging Handwritten Document Images Using Fully Convolutional Network

This paper presents a method for text line segmentation of challenging\nhistorical manuscript images. These manuscript images contain narrow interline\nspaces with touching components, interpenetrating vowel signs and inconsistent\nfont types and sizes. In addition, they contain curved, multi-skewed and\nmulti-directed side note lines within a complex page layout. Therefore,\nbounding polygon labeling would be very difficult and time consuming. Instead\nwe rely on line masks that connect the components on the same text line. Then\nthese line masks are predicted using a Fully Convolutional Network (FCN). In\nthe literature, FCN has been successfully used for text line segmentation of\nregular handwritten document images. The present paper shows that FCN is useful\nwith challenging manuscript images as well. Using a new evaluation metric that\nis sensitive to over segmentation as well as under segmentation, testing\nresults on a publicly available challenging handwritten dataset are comparable\nwith the results of a previous work on the same dataset.\n