Face comparison method based on high-dimensional LBP (Local Binary Patterns) and convolutional neural network feature fusion

Face comparison method based on high-dimensional LBP (Local Binary Patterns) and convolutional neural network feature fusion

  • CN 105,550,658 A
  • Filed: 12/24/2015
  • Published: 05/04/2016
  • Est. Priority Date: 12/24/2015
  • Status: Active Application
First Claim
Patent Images

1. , based on a face comparison method for higher-dimension LBP and convolutional neural networks Fusion Features, it is characterized in that comprising following four steps:

  • Step (1), first will human face region be detected and intercept human face region image from the image of input, and then facial image be alignd, zoom to specific dimensions, finally image is carried out histogram equalization;

    Step (2), using the facial image after step (1) histogram equalization as input, the higher-dimension LBP feature and the convolutional neural networks feature that obtain facial image respectively carry out splicing the proper vector obtaining 6096 dimensions, carry out dimensionality reduction by PCA again, obtain the proper vector of 1024 dimensions;

    Step (3), two facial images are all through step (1) and step (2), obtain the proper vector of two 1024 dimensions, as input, (combine Bayesian model by the JointBayesian trained, obtain corresponding log-likelihood ratio;

    The threshold value of step (4), log-likelihood ratio step (3) obtained and priori is compared, if log-likelihood is larger than threshold value, then thinks that these two facial images are same person, otherwise is not same person.

View all claims
    ×
    ×

    Thank you for your feedback

    ×
    ×