Effect of normalization on activation

I know that normalization tends to center the data making the learning process more stable.
But what is it’s effect on the intermediate activation/feature maps.
for example below is an image before normalization along with it’s histogram


histogram of image
berlin_hist
Same image after histogram equalization is done along with it’s histogram

image histogram after equalization is done
berlin_hist_after
clearly we can see that after the equalization is done,the image looks more dynamic,where you can clearly see the buildings.

so i am thinking that if normalization wasn’t done,then different parts in the image would end up firing differently,thereby making it difficult for it to learn features and discriminate between them. am i right in thinking like this?