how to reduce the dimension of a feature space?

I have 260 sample,each sample has 320 feature (x is a matrix with 260 rows(images) & 320 column(features)).in order to improve my classification,i need to reduce these 320 column(i mean number of features).but i dont know how to do. when i use for example:

[pc,score,latent,tsquare] = princomp(X); red_dim = score(:,1:50); how to reconstruct the matrix with fewer column?

when i use :

residuals = pcares(X,ndim) the dimension of residuals is the same of x !!!!

actually i dont know what is my new matrix?

[Walter Roberson]

duplicates http://www.mathworks.co.uk/matlabcentral/answers/78042-how-to-reduce-the-dimension-of-a-feature-space