How Can I Combine PCA Models for Enhanced Face Recognition?

[MrDaniel]

Hello,

I am working with face recognition. I have two models from two separate datasets with the same number of dimensions.

I wish to implement this method to combine PCA models.

http://www.cs.cf.ac.uk/Dave/Papers/Pami-EIgenspace.pdf

My linear algebra isn't great. So i am lost after step 1 for section 3.1.

I can appreciate how to get the combined mean, but from that point on wards, it's a bit over my head.

Anyone here literate in this symbolic medium of communication?

 

[jvicens]

Hello,

I am a scientist with experience in face recognition and PCA (Principal Component Analysis). I understand that you are trying to combine two PCA models from different datasets. The method you have mentioned in your post is a commonly used technique called Eigenface method.

After step 1 in section 3.1, you need to calculate the covariance matrices for both datasets. This can be done using the formula COV = (X-X_mean) * (X-X_mean)^T, where X is the dataset and X_mean is the mean of the dataset. This will give you two covariance matrices, one for each dataset.

Next, you need to calculate the eigenvectors and eigenvalues for each covariance matrix. These can be obtained using a mathematical tool called SVD (Singular Value Decomposition). The eigenvectors will represent the principal components of each dataset.

Now, you need to combine the eigenvectors from both datasets by stacking them on top of each other. This will give you a new matrix with the combined eigenvectors.

Finally, you can use this combined eigenvector matrix to project your data onto a lower dimensional space, reducing the number of dimensions while preserving the most important features of the data.

I hope this explanation helps you understand the steps after step 1 in section 3.1. If you have any further questions, please feel free to ask. I am happy to help.