While Merging image datasets which of the image parameters should be prepossessed/Normalized before giving to a CNN Model?

When two datasets are merged or images of different parameters size, dimension, Format are combine which parameters of the datasets should be normalized/ pre-processed before giving it to a model?