2016/12/22 by Bruno Stuner, Clément Chatelain, Stuner, Bruno +3
Computer Science · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Handwritten Text Recognition Techniques #Natural Language Processing Techniques #Text and Document Classification Technologies
paper · pdf · doi:10.48550/arxiv.1612.07528
openalex publication_date 2016/12/22 · openalex created_date 2022/10/02 · openalex updated_date 2026/07/28
State-of-the-art methods for handwriting recognition are based on Long Short\nTerm Memory (LSTM) recurrent neural networks (RNN), which now provides very\nimpressive character recognition performance. The character recognition is\ngenerally coupled with a lexicon driven decoding process which integrates\ndictionaries. Unfortunately these dictionaries are limited to hundred of\nthousands words for the best systems, which prevent from having a good language\ncoverage, and therefore limit the global recognition performance. In this\narticle, we propose an alternative to the lexicon driven decoding process based\non a lexicon verification process, coupled with an original cascade\narchitecture. The cascade is made of a large number of complementary networks\nextracted from a single training (called cohort), making the learning process\nvery light. The proposed method achieves new state-of-the art word recognition\nperformance on the Rimes and IAM databases. Dealing with gigantic lexicon of 3\nmillions words, the methods also demonstrates interesting performance with a\nfast decision stage.\n