How to add noise to every image in image datastore?

Hi! I am trying to add gauss noise to every image in image datastore. I want to compare a network trained on normal images with a network trained on only noisy images. A found a soltion here:

https://www.mathworks.com/help/deeplearning/ug/image-to-image-regression-using-deep-learning.html

The way the noise is added here, is by creating a TransformedDatastore, but I can not use it to train a netvork. It gives the folowing error:

It also threw an error every time I wanted to use it's own object functions, even though I followed the steps in the link above. I basically want to have two image datastores with the same pictures, one with noise, one without noise. Is there a way to do that?

Answer 1

[Shadaab Siddiqie]

From my understanding you want add gaussian noise to all the image, you can use imnoise and loop through all the images from your dataset.

[Csanad Levente Balogh]

Thank you! Yes I assumed it can be done with loops, I just wanted to find a more elegant way. At the end, I used the custom ReadFcn function of the datastore to add the noise. like that:

imageDatastore.ReadFcn = @customReadDatastoreImage

Than the function:

function data = customReadDatastoreImage(filename)
% other settings
% ...

data = imread(filename);

% other preproc (like imresize)
% ...

data = imnoise(data, 'gaussian')

end