Generate new image from principal components of many images

Hey everyone. I'm having a bit of difficulty with a neuroscience course - it's about the primary visual cortex.

I have a bunch of image patches (16*16) and I need to generate synthesised images using principal component analysis as a generative model. I reshaped the images to vectors with 256 elements (50000 observations) and got the principal components (a 256*256 matrix)

Now I need to 'generate synthesized images using my PCA as a generative model and assuming that the marginal distribution of the components is Gaussian with a variance equal to the variance of the learned component.'

I've thought about it for ages, and I just can't understand what the assignment is trying to get me to do. Could anyone please please please offer me some help? I'm really stuck.
Last edited: