Principal Component AnalysisDr. Nidhi Mathur
Principal Component Analysis or PCA is a way of identifying patterns in data and expressing data in such a way as to highlight their similarities and differences.PCA is a powerful tool for analyzing data.2PCA - Dr. Nidhi  Mathur
MethodGet some data,Subtract the mean,Calculate the covariance matrix,Calculate the eigenvectors and eigenvalues of covariance matrix,Choose components and form a feature vector,Derive the new data set.3PCA - Dr. Nidhi  Mathur
Image RepresentationRows  of the pixels in an (NxN) image are placedone after the other to form a one dimensional vector.N x N Image1 x N2 vector4PCA - Dr. Nidhi  Mathur
If there are M images, thenNow, this is the starting point of PCA Analysis.5PCA - Dr. Nidhi  Mathur
Let  be an 1 x N 2 vector corresponding to an N x N image.Obtain images I1, I2, …..IM.Represent every Ii as vector i.Compute  average vector Subtract the mean vectorCompute the covariance matrix ,        where                            is an N 2 x M matrixCompute the eigenvalues  ui of ATA6PCA - Dr. Nidhi  Mathur
Matrix ATA is very large and  computation is impractical.Consider AAT matrix.
Compute  eigenvalues vi of AAT. (AT A and AAT have the same eigenvalues and their eigenvectors are related as: ui = A x vi ) AT A has  N2 eigenvectors/eigenvalues
AAT  has M eigenvectors/eigenvalues7PCA - Dr. Nidhi  Mathur

PCA

  • 1.
  • 2.
    Principal Component Analysisor PCA is a way of identifying patterns in data and expressing data in such a way as to highlight their similarities and differences.PCA is a powerful tool for analyzing data.2PCA - Dr. Nidhi Mathur
  • 3.
    MethodGet some data,Subtractthe mean,Calculate the covariance matrix,Calculate the eigenvectors and eigenvalues of covariance matrix,Choose components and form a feature vector,Derive the new data set.3PCA - Dr. Nidhi Mathur
  • 4.
    Image RepresentationRows of the pixels in an (NxN) image are placedone after the other to form a one dimensional vector.N x N Image1 x N2 vector4PCA - Dr. Nidhi Mathur
  • 5.
    If there areM images, thenNow, this is the starting point of PCA Analysis.5PCA - Dr. Nidhi Mathur
  • 6.
    Let  bean 1 x N 2 vector corresponding to an N x N image.Obtain images I1, I2, …..IM.Represent every Ii as vector i.Compute average vector Subtract the mean vectorCompute the covariance matrix , where is an N 2 x M matrixCompute the eigenvalues ui of ATA6PCA - Dr. Nidhi Mathur
  • 7.
    Matrix ATA isvery large and computation is impractical.Consider AAT matrix.
  • 8.
    Compute eigenvaluesvi of AAT. (AT A and AAT have the same eigenvalues and their eigenvectors are related as: ui = A x vi ) AT A has N2 eigenvectors/eigenvalues
  • 9.
    AAT hasM eigenvectors/eigenvalues7PCA - Dr. Nidhi Mathur