Feature dimension reduction using PCA

Hello everyone, I am trying to reduce feature dimensions using PCA. The feature matrix is 12614x1536 where 12614 are images and 1536 are features. For PCA I am using the following code:

[coeff, score] = pca(feature_matrix);
new_features = score(:,1:400) * coeff(:,1:400)';

Here, I want to select 400 features but the new_features dimensions are the same as the old features 12614x1536. It should be 12614x400. Please need your feedback, seems I am missing something... Thank you.

Answer 1

[Image Analyst]

According to the help, this is the formula:

[coeff1,score1,latent,tsquared,explained,mu1] = pca(y,'algorithm','als');
% Reconstruct the observed data.
t = score1*coeff1' + repmat(mu1,13,1)

And it will just reconstruct the original data, not give you new features.