Handwritten Text Recognition Using Deep Learning

Abstract

Handwritten textual content reputation (HTR) represents an ongoing subject undertaking due to the fact distinctive writing patterns blended with more than one handwriting styles create enormous reputation complexity. The paper offers a brand new HTR answer that mixes Vision Transformers with MobileNet-LSTM hybrid structure to maximise reputation precision and operational speed. The Vision Transformer demonstrates excellence at shooting long-distance picture relationships whilst extracting significant traits from handwritten information. MobileNet serves to reduce computational necessities of the device even as maintaining excessive overall performance requirements and the device's temporal dependency ability has been executed thru Long Short-Term Memory (LSTM) community utility for series modeling in handwriting. Engineered from convolutional neural community (CNN)-primarily based totally processes the proposed assessment technique affords higher reputation quotes on standardized benchmark datasets. Experimental information confirms the ViT and MobileNet-LSTM aggregate solves real-time handwritten textual content reputation troubles via way of means of accomplishing excessive accuracy quotes with green computational necessities.