Handwritten Text Recognition using Deep Learning and Word Beam Search

This paper offers a solution to traditional handwriting recognition techniques using concepts of Deep learning and Word Beam Search. explains about how an individual handwritten word is classified from the text by translating into digital form. The form when trained with Connectionist Temporal Classification (CTC) loss function, output produced RNN. matrix containing character probabilities for each time-step. final mapped CTC decoding algorithm converting probabilities. recognized constructed list words dictionary token passing algorithm. It found running time depends on size dictionary. Also numbers like arbitrary strings will not able decode. In this search beam proposed, in order tackle these types problems. methodology support constrain similar those contained allows such as non-word between words, integrates word-level language model. better compared passing. proposed comprises named vanilla IAM dataset Bentham data set.