Dissertation > Excellent graduate degree dissertation topics show

Manifold Structure Analysis and Recognition for Signer-Independent Sign Language Data

Author: XuRong
Tutor: YaoHongXun
School: Harbin Institute of Technology
Course: Computer Science and Technology
Keywords: Non - specific staffing language recognition Hidden Markov Models Manifold Learning Isomap Tangent vector
CLC: TP391.41
Type: Master's thesis
Year: 2007
Downloads: 62
Quote: 1
Read: Download Dissertation

Abstract


The purpose of the sign language recognition research is to enhance barrier-free exchanges between the Deaf and hearing health, improve your computer's ability to understand human language. The non-specific staffing language recognition is to promote the practical use of sign language recognition system must solve the problem. , Non-specific staffing language recognition and specific staffing language recognition performance, there is still a large gap, mainly due to the difference of the data itself contradictions and lack of training samples. The data discrepancies contradictions makes non-specific staffing language recognition to extract data valid common feature sign language is very difficult. In practical applications, the model of the contradiction between the lack of expression ability with the sample has become a bottleneck Restriction recognition system effects. To solve the above two issues, this article will be manifold ideas incorporated into traditional HMM model of sign language, the main research work are as follows: 1. Isomap algorithm intuitive manifold structure of the data to show sign language. Similar data sets in geometry a nested manifold structure having a certain formed by intrinsic invariance, the structure itself corresponds to the concept of a single manifolds. Isomap method is the most common manifolds visualization algorithm, the data processing result of the sign language sign language data itself contains a manifold structure, and may be corresponding to the HMM model in the state classes and manifold concept mining its intrinsic invariance. Visualization according Isomap result, proposed a TV / HMM sign language model. The manifold concept has the advantage of some learning and reasoning ability, the variability of the tangent vector to a linear expression data for effective modeling, the degree of change allowed in the class, so that the classifier of some category changes caused data not sensitive to changes, and obtained through a description of the geometry, in the case of the smaller training set, formalized description of the data of individual differences. TV / HMM model both added processing of specific factors, to make up for the lack of training data defects. 3 hand TV / HMM-based word recognition system. The system uses maximum likelihood estimation tangent vector learning from the training data, to determine the optimal parameters and the number of iterations, TV / HMM model relative to the superiority of the traditional HMM model is proved by experiments. In the training set is small, does not significantly increase the time complexity of the case, TV / HMM model is a great improvement on the system performance may be the identification of the non-specific, from 70.38% to 72.44%, the degree of improvement of 6.96% ; join virtual data generated within the mean-shift classes to the non-specific identification rate of 70.56% to 72.94%, the degree of improvement of 8.07%.

Related Dissertations

  1. The Classification of High Dimsnsion Flew Field Based on Manifold Learning,V231.3
  2. Research and Implementation of Time Series Classification Based on Semi-supervised Learning,TP181
  3. Based on multi- location basis ISOMAP Algorithm Research,TP181
  4. Incremental Spectral Embedding Methods in Manifold Learning,TP181
  5. Facial expression recognition research based on manifold learning,TP391.41
  6. Robust manifold learning algorithm and its application,TP181
  7. Reseach on Nonlinear Dimensionality Reduction Methods in Manifold Learning and Their Applications for Tobacco Data,TP181
  8. The robustness Research manifold learning,TP181
  9. Semi-Supervised Manifold Learning Theory and Applications,TP181
  10. Based on manifold learning Chinese Web Text Classification Algorithm,TP391.1
  11. Age estimation of face images,TP391.41
  12. Research on the Method of Knowledge Acquisition Based on Concept Lattice and Manifold Learning,TP182
  13. Theories and Algorithms of Manifold Learning and Applications in Biometric Authentication,TP181
  14. Optimization and Application Research of Neural Network Based on PSO Algorithm,TP183
  15. The Cough Signal Spotting in Continuous Speech,TN912.34
  16. Manifold Learning on Probabilistic Graphical Models,TP181
  17. Improvement and Implementation of Manifold Learning Algorithms ISOMAP,TP181
  18. Video Analysis Using Semi-supervised Manifold Learning,TP391.41
  19. Feature Extraction, Selection and Combination in Lipreading,TP391.41
  20. The Analyzing and Application of Dimensionality Reduction Based on Manifold Learning and Subspace,TP391.41
  21. Research on Sample Set Construction and Personalized Generation for 3D Cartoon Face,TP391.41

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net  Mobile