Unconstrained off-line handwriting text recognition in general and for Arabic-like scripts in particular is a challenging task and is still an active researcharea. Transformer based models for English handwriting recognition have recently shown promising results.In this paper, we have explored the use of transformerarchitecture for Urdu handwriting recognition. The useof a Convolution Neural Network before a vanilla fullTransformer and using Urdu printed text-lines alongwith handwritten text lines during the training are thehighlights of the proposed work. The Convolution Layers act to reduce the spatial resolutions and compensate for the n2 complexity of transformer multi-head attention layers. Moreover, the printed text images inthe training phase help the model in learning a greaternumber of ligatures (a prominent feature of Arabiclike scripts) and a better language model. Our modelachieved state-of-the-art accuracy (CER of 5.31%) onpublicly available NUST-UHWR dataset [1].

Unconstrained off-line handwriting text recognition in general and for Arabic-like scripts in particular is a challenging task and is still an active research area. Transformer based models for English handwriting recognition have recently shown promising results. In this paper, we have explored the use of transformer architecture for Urdu handwriting recognition. The use of a Convolution Neural Network before a vanilla full Transformer and using Urdu printed text-lines along with handwritten text lines during the training are the highlights of the proposed work. The Convolution Layers act to reduce the spatial resolutions and compensate for the n 2 complexity of transformer multi-head This research is partially funded by Higher Commission (HEC), Pakistan's grant for the National Center of Artificial Intelligence (NCAI).