Dissertation > Excellent graduate degree dissertation topics show

Based on the the the flow -shaped learning and the the multi- of the the Zhang the amount of of the pose face recognition research

Author: LiWeiQing
Tutor: ChenDuanSheng
School: Huaqiao University
Course: Applied Computer Technology
Keywords: Face Recognition Pose estimation Manifold learning Tensor face
CLC: TP391.41
Type: Master's thesis
Year: 2011
Downloads: 85
Quote: 0
Read: Download Dissertation

Abstract


In face recognition in the, the the If the the people face image is a non-the front of is, identification the effect of will be greatly reduce the. If the it is pre-estimated that the The the posture of the human face for a further select the the a suitable of the viewing angle model of to to carried out the identification, will be improve the the the the recognition rate of of the non--a positive human face. The method of the tensor an image of put human face as is the comprehensive the results of of the multi-factors of the of the geometric structure, The posture and an illumination, and, etc., and can be separation to out of the space of the the various sub-. The stream-shaped learning methods can be will be the the posture sub-space are arranged in a the one continuous the the posture of stream shaped, and be able to estimate for out the the the posture of approximation of of the new Shipment a. Consider the, In research when Face Recognition Based on Zhang amount of the multi-pose, the the first with a Manifolds learning is estimated to be out the the scope of the the posture of of the human face, and then will be identity the sub-tensor At A to invest in the the posture of in the within this range only conducted identification. This way have a greater changes in on the in the human Face Pose when the, also able to get a better recognition rate. According to the This train of thought, the of this article has to carry out the following research the work of: 1. Compare based on the estimation method of the Face Pose of the the people who learning by the stream-shaped: the learning of the people stream-based-shaped the estimation method of the Face Pose there are: LLE, LE, LEA, LPP and other, of which LEA, LPP , respectively, is the the the approximate linear improved method for of of LLE, LE. The estimated of the method of is selected by one kinds estimation error and small method as a the of this article research posture, the of this article By comparing the experimental the research methods for the, the choice of the LEA as a the estimated by the of this article posture. 2. On the the LEA posture estimated method to do has been improved: LEA in the the the the when the the estimated of the people Face Pose, select the the the partial ortho domain only consider the of similar posture, this way on the lose a adjacent posture of the geometric topology information. Of this article put forward the select method of the the An Improved the ortho domain, take full advantage of the a priori attitude information, so that the of similar posture with each other close to the, of heterogeneous posture distance With the the the posture the difference variable large while the increases, and the be able to to so that the the low-dimensional of the the training and test of sample of the stream-shaped The even more near, the reduce the error. The in the the on the the database of Facepix human face of the experimental prove that the a the the validity of of the of the method. 3. Asked the one kinds the based on the the of the multi-posture of of the the stream-shaped learning and the tensor face Face Recognition Method: the the the the traditional tensor face method when the the face recognition is in progress, the there are are the following disadvantages: (1) will be the sample to be identification investors to invest in all posture, calculation large amount of low recognition rate; (2) use the to the tensor 's face said that people face when-dimensional the number of very high, traditional in order to the PCA for the dimensionality reduction, but the PCA is the linear methods, while the the the tensor' s face is of higher order nonlinear structure, dimensionality reduction after the will be lost the non-linear information. In response to these disadvantage of, First of all use the that this article proposed by to. This article improvements the LEA method estimated the people face posture the scope of, when the the identification of in the tensor 's face only invest in the posture in the in the within the scope of, and both to improve the the efficiency of to and also enhance the the recognition rate; Secondly use LLE instead of PCA carried out tensor dimensionality reduction, LLE is nonlinear dimensionality reduction method, be able to maintain the the non-linear structure in the when the the high-dimensional after the the dimensionality reduction. The the of this article method is very good of the to solve the the the the the the problem of insufficient of the the non-linear processing of the Zhang the amount of, in the the when the posture changes in the the larger, but also be able to compare good carried out the face recognition. Large number of experiments to prove a the the validity of of the this method.

Related Dissertations

  1. The Classification of High Dimsnsion Flew Field Based on Manifold Learning,V231.3
  2. Research on Algorithms of 2D Face Template Protection,TP391.41
  3. Feature Extraction, Selection and Combination in Lipreading,TP391.41
  4. Study of Data Reduction Technique Based on Manifold Learning,TP311.13
  5. Research of Video Face Recognition Based on Weighted Voting and Key-Frame Extraction,TP391.41
  6. Face Recognition Method Based on DE,TP391.41
  7. A Novel Approach for Projection by Optimize Local Area in Face Recognition Applications,TP391.41
  8. Research on Deep Structure Learning Algorithms Based on Dynamic Fuzzy Relation,TP181
  9. The Research for Face Recognition Based on Pseudo-Zernike Moment and BP Network,TP391.41
  10. The image -based face recognition system design and implementation beautification,TP391.41
  11. Fusion depth map and 3D models of human motion capture technology research,TP391.41
  12. Based on wavelet transform and linear subspace face recognition technology,TP391.41
  13. \,R338
  14. Expression of face recognition algorithm based on sparse research,TP391.41
  15. Time series visualization of human motion and multi- index method,TP391.41
  16. Research on Subspace Analysis Based Face Recognition Algorithm,TP391.41
  17. Research on Face Recognition Based on Improved Maximum Margin Criterion,TP391.41
  18. Research on Key Technology of LTE Intrusion Detection Systems,TP393.08
  19. Research on the Technology of Age Estimation Based on Gender Identification,TP391.41
  20. Research of Face Recognition Technology,TP391.41
  21. Face Recognition Method Base on Quaternion Wavelet Magnitude/Phase Features,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