From Recurrent Neural Network to Long Short Term Memory Architecture
Résumé
Despite more than 30 years of handwriting recognition research, Recognizing the unconstrained sequence is still a challenge task. The difficulty of segmenting cursive script has led to the low recognition rate. Hidden Markov Models (HMMs) are considered as state-of-theart methods for performing non-constrained handwriting recognition. However, HMMs have several well-known drawbacks. One of these is that they assume the probability of each observation depends only on the current state, which makes contextual effects difficult to model. Another is that HMMs are generative, while discriminative models generally give better performance in labelling and classification tasks. Recurrent neural networks (RNNs) do not suffer from these limitations, and would therefore seem a promising alternative to HMMs. A novel type of recurrent neural network, termed as Bidirectional Long Short-Term Memory (BLSTM) architecture, will be studied in this thesis. A sequence concatenating technique called Connecionist Temporal Classification (CTC) is applied. Finally, Three extended decoding algorithm: Levenshtein Distance(LD), full path(FD), max path(MD) are proposed insighted by HMM to have a lexicon-based classification. The system BLSTM-CTC-FP is demonstrated to be robust to lexicon-based recognition and reduce 50% error than the existing best model.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...