Keras ImageDataGenerator allows feature normalization as below:
I am working with Diabteic Retinopathy data , which are medical images. I have following question.
Feature normalization is supposed to help with features being at different scale during training a CNN/Neural network.
For medical images like in above dataset, does feature normalization (mean subtraction and dividing by standard deviation of feature set)help with color balance and or brightness adjustment in the image or contrast improvement?
Does Keras ImageDataGenerator help with this? my question is that using algorithm like Retinex algorithm or histogram equalization for contrast improvement and then doing feature normalization within keras ImageDataGenerator framework is a correct approach or feature normalization is sufficient?