Comparative Study of HMM and BLSTM Segmentation-Free Approaches for the Recognition of Handwritten Text-Lines

Abstract

This paper deals with the recognition of freestyle handwritten text lines. We compare 2 state-of-the-art segmentation-free recognition approaches. The first one is the popular context-dependent HMM approach (Hidden Markov Models). The second one is the recent BLSTM (Bi-directional Long Short-Term Memory) approach based on recurrent neural networks and memory blocks. For the sake of comparison, both recognizers use the same set of features and language model. They are compared from the following perspectives: sliding window parameters for feature extraction, training and decoding speed and performance accuracy with or without using a language model. We compare these two approaches on the publicly available Rimes database of French handwritten mails. Our main findings are that long frame sequences, obtained with specific window parameters, improve both recognizers, and that BLSTMs outperform HMMs in terms of WER rates, at the expense of considerably longer training times.

Document Analysis and Recognition: A survey

“…Other NN architectures proposed are encoder-decoder with attention, Gated FCN, Gated CNN, The DAR tasks with the use of NNs in the literature are character recognition [169], [184] [213] survey of NN approaches [406], word recognition, line recognition [188], attention based recognition, WI [31], OD models for recognition [115], etc. RNN overtook the road of recognition problems with the development of deep networks.…”

“…Figure 7 a) shows the character and word HMM model and various combinations of HMM with NNs. Some hybrid models of HMM like ANN/HMM [187], CNN/HMM [181], HMM/BLSTM [188] model is also proposed for improved recognition performance. HMM, models can be used for end-to-end recognition [189] and online recognition [190].…”

“…HMM, models can be used for end-to-end recognition [189] and online recognition [190]. Other tasks of DAR where the use of HMM has been proposed are Character classification, character recognition [183], word recognition [151], [159], [181], [191], [192], line recognition [188], numeral recognition [193] and text recognition [179].…”

“…Other techniques of end-to-endrecognition systems are discussed below: a) Hybrid Modelling: The early methods used mostly HMMs for end-to-end recognition, but HMMs were limited in modelling the long-term dependencies [189]. Also, some combination models like HMM-ANN [187], HMM-CNN [181], HMM-BLSTM ANN/HMM and CNN/HMM [188] were proposed to deal with the shortcomings of the former approach. The recent methods using only CNNs have improved accuracy parameters like recall, precision, etc.…”