Unsupervised learning of text line segmentation by differentiating coarse patterns

Despite recent advances in the field of supervised deep learning for text\nline segmentation, unsupervised deep learning solutions are beginning to gain\npopularity. In this paper, we present an unsupervised deep learning method that\nembeds document image patches to a compact Euclidean space where distances\ncorrespond to a coarse text line pattern similarity. Once this space has been\nproduced, text line segmentation can be easily implemented using standard\ntechniques with the embedded feature vectors. To train the model, we extract\nrandom pairs of document image patches with the assumption that neighbour\npatches contain a similar coarse trend of text lines, whereas if one of them is\nrotated, they contain different coarse trends of text lines. Doing well on this\ntask requires the model to learn to recognize the text lines and their salient\nparts. The benefit of our approach is zero manual labelling effort. We evaluate\nthe method qualitatively and quantitatively on several variants of text line\nsegmentation datasets to demonstrate its effectivity.\n