Pho(SC)-CTC—a hybrid approach towards zero-shot word image recognition

Annotating words in a historical document image archive for word recognition purpose demands time and skilled human resource (like historians, paleographers). In real-life scenario, obtaining sample images all possible is also not feasible. However, zero-shot learning methods could aptly be used to recognize unseen/out-of-lexicon such images. Based on previous state-of-the-art method “Pho(SC)Net”, we propose hybrid model based the CTC framework (Pho(SC)-CTC) that takes advantage of rich features learned by Pho(SC)Net followed “connectionist temporal classification” (CTC) perform final classification. Encouraging results were obtained two publicly available datasets one synthetic handwritten dataset, which justifies efficacy Pho(SC)-CTC Pho(SC)Net.