Loading...

Automatic Concept Extraction to Improve the Recognition Performance for Sequential Patterns

Halavati, Ramin | 2009

524 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 40104 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Bagheri Shouraki, Saeid
  7. Abstract:
  8. In this dissertation, we introduced a Fuzzy based representation and comparison method for sequential patterns such as speech and online handwriting. The new model, called Fuzzy Elastic Matching Machine (FEMM), is simpler than traditional HMM based approaches and is not based on the common statistical assumptions of HMM systems. The model was tested on isolated word and phoneme recognition tasks in speech recognition domain and isolated letter recognition in Persian handwriting recognition. We showed that this method is faster than traditional HMM based models and more robust to noise. To train the model, we introduced a Symbiogenesis-based evolutionary training algorithm. This algorithm can be used as a general purpose knowledge extraction method for data mining and we showed that while having at least similar classification results in compare with traditional genetic based methods, it has much faster convergence speed. We also introduced a special purpose version of this algorithm for FEMM training task resulting in much faster training in compare with hill climbing method which is quite common for MMI or MCE training tasks in HMMs. Also we showed that using this algorithm, adding new classes to a previously trained classifier requires less training time in compare to when all classes are trained together as the training algorithm can make use of its formerly generated partial descriptions. Also to improve the recognition speed, we presented a pre-clustering algorithm which compresses the data stream before passing it to recognition models. We showed that this algorithm can reduce the size of data by a factor of at least two and half without loss of recognition accuracy resulting in more than 60% faster recognition.

  9. Keywords:
  10. Pattern Recognition ; Fuzzy Modeling ; Machine Learning ; Speech Recognition ; Handwriting Recognition ; Symbiotic Evolution

 Digital Object List

 Bookmark

No TOC