Changing the distribution of characters in words dataset to uniform distribution

Question

I have dataset of text images, each image contain 1 to 3 words. I need to predicate the sequence of characters in these images, however the distribution is very skewed. I found