## Offline Cursive Word Recognition using Continuous Density Hidden Markov Models trained with PCA or ICA Features (2001)

### Cached

### Download Links

Venue: | ICA Features,” Sixth International Conference on Pattern Recognition (ICPR 2002 |

Citations: | 10 - 3 self |

### Citations

6474 | Neural Networks for Pattern Recognition - Bishop - 1995 |

5745 | A tutorial on hidden markov models and selected applications in speech recognition
- Rabiner
- 1989
(Show Context)
Citation Context ...ell i. The feature vector collects the values f i = n i P j n j . 3. Hidden Markov Models Hidden Markov Models are probability density functions over sequences of vectors (for a good introduction see =-=[4]-=-). The sequences are assumed to be produced by a system characterized by a state (belonging to a finite set of possible states Q = fQ i : i = 1; 2; : : : ; Ng) that changes at discrete time steps. The... |

2236 | Independent Component Analysis
- Hyvärinen, Karhunen, et al.
- 2001
(Show Context)
Citation Context ...f W determines the properties of y. We used two different criteria leading to the extraction of the Principal Components (using PCA) and of the Independent Components (using ICA). When performing PCA =-=[2]-=-, the rows of W are the eigenvectors of the covariance matrix of the original data (assumed to have 0 mean, condition that can always be easily achieved by substracting the mean estimated over the tra... |

1805 | Independent component analysis: A new concept - Comon - 1994 |

1458 | Gradient-based learning applied to document recognition - LeCun, Bottou, et al. - 1998 |

1150 | Cross-validatory choice and assessment of statistical predictions - Stone - 1974 |

1113 | A maximization technique occurring in the statistical analysis of probabilistic functions of Markov Chains - Baum, Petrie, et al. - 1970 |

982 | The viterbi algorithm - Forney - 1973 |

913 | Statistical Methods for Speech Recognition - Jelinek - 1997 |

621 | An inequality and associated maximation technique in statistical estimation for probabilistic functions of markov processes - Baum - 1972 |

594 |
Computer and Robot Vision
- Haralik, Shapiro
- 1992
(Show Context)
Citation Context ...f all the elements in the word image that are not useful for the recognition. The operations performed at this stage depend on the data. In our case, a binarization (performed with the Otsu algorithm =-=[1]-=-) is sufficient. The normalization is supposed to remove slant (the angle between the vertical direction and the direction of the strokes supposed to be vertical in an ideal model of handwriting) and ... |

537 | Statistical inference for probabilistic functions of finite state markov chains - Baum, Petrie - 1966 |

282 | An inequality with applications to statistical estimation for probabilistic functions of Markov processes and to a model for ecology - Baum, Eagon - 1967 |

269 | Kernel principal component analysis - Schölkopf, Smola, et al. - 1999 |

257 | Feature extraction methods for character recognition-a survey - Trier, Jain, et al. - 1996 |

127 | H.: Using a statistical language model to improve the performance of an hmm-based cursive handwriting recognition system - Marti, Bunke - 2001 |

116 | Markovian models for sequential data - Bengio - 1999 |

106 | Off-line Cursive Script Word Recognition - Bozinovic, Srihari - 1989 |

88 |
An Off-Line Cursive Handwriting Recognition System
- Senior, Robinson
- 1998
(Show Context)
Citation Context ...e raw data). The best system was obtained by training over the Principal Components. Its performance over the data set used is significantly higher than the accuracies claimed (over the same data) in =-=[5, 3]-=- (92:8% and 85:0% respectively) and slightly better than the 94:6% recognition rate presented in [6]. On the other hand, this last system is much more complex than our. The words are first segmented i... |

55 | Growth functions for transformations on manifolds - Baum, Sell - 1968 |

28 | A fast algorithm for independent component analysis - Hyvarinen, Oja - 1997 |

22 | A new normalization technique for cursive handwritten words
- Vinciarelli, Luettin
(Show Context)
Citation Context ...of a reasonable interval) and the shear transformed image giving the highest value of deslantedness is assumed as the deslanted one. For a full description of the normalization technique we used, see =-=[7]-=-. The slope and slant removal methods applied are adaptive and do not use any parameters to be set empirically. This avoids the need of tuning a different parameter set for each writer and makes the s... |

14 | Fast and robust algorithms for independent component analysis - Hyvarinen - 1999 |

13 | On-line and o-line handwriting recognition: A comprehensive survey - Plamondon, Srihari |

12 |
Towards general cursive script recognition
- Marti, Bunke
- 1999
(Show Context)
Citation Context ...e raw data). The best system was obtained by training over the Principal Components. Its performance over the data set used is significantly higher than the accuracies claimed (over the same data) in =-=[5, 3]-=- (92:8% and 85:0% respectively) and slightly better than the 94:6% recognition rate presented in [6]. On the other hand, this last system is much more complex than our. The words are first segmented i... |

12 | Intrator,“Off-Line Cursive Script Word Recognition A Survey - Steinherz, Rivlin, et al. |

9 | Independent component analysis: A tutorial. Neural Networks - Hyvärinen, Oja - 2000 |

7 | An offline cursive handwritten word recognition system
- Tay, Lallican, et al.
- 2001
(Show Context)
Citation Context ...er the data set used is significantly higher than the accuracies claimed (over the same data) in [5, 3] (92:8% and 85:0% respectively) and slightly better than the 94:6% recognition rate presented in =-=[6]-=-. On the other hand, this last system is much more complex than our. The words are first segmented into primitives using a sliding window approach. A Neural Network is then used 2 4 6 8 10 12 14 85 90... |

6 | Error bounds for convolutional codes and an asimptotically optimal decoding algorithm - Viterbi - 1967 |

3 | Offline Handwritten Word Recognition Using A Hybrid Neural Network And - Tay’, hllicad, et al. |